]>
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 1.54 2001/08/16 15:24:21 stelian Exp $";
48 #include <compatlfs.h>
54 #include <compaterr.h>
65 #include <sys/types.h>
69 #include <sys/param.h>
70 #include <sys/socket.h>
75 #ifdef HAVE_EXT2FS_EXT2_FS_H
76 #include <ext2fs/ext2_fs.h>
78 #include <linux/ext2_fs.h>
80 #include <ext2fs/ext2fs.h>
81 #include <bsdcompat.h>
83 #include <sys/vnode.h>
86 #include <ufs/inode.h>
88 #include <ufs/ufs/dinode.h>
89 #include <ufs/ffs/fs.h>
90 #endif /* __linux__ */
92 #include <protocols/dumprestore.h>
96 #endif /* HAVE_ZLIB */
100 #endif /* HAVE_BZLIB */
104 int writesize
; /* size of malloc()ed buffer for tape */
105 long lastspclrec
= -1; /* tape block number of last written header */
106 int trecno
= 0; /* next record to write in current block */
107 extern long blocksperfile
; /* number of blocks per output file */
108 long blocksthisvol
; /* number of blocks on current output file */
109 extern int ntrec
; /* blocking factor on tape */
110 extern int cartridge
;
115 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
117 static ssize_t atomic_read
__P((int, void *, size_t));
118 static ssize_t atomic_write
__P((int, const void *, size_t));
119 static void doslave
__P((int, int, int));
120 static void enslave
__P((void));
121 static void flushtape
__P((void));
122 static void killall
__P((void));
123 static void rollforward
__P((void));
126 * Concurrent dump mods (Caltech) - disk block reading and tape writing
127 * are exported to several slave processes. While one slave writes the
128 * tape, the others read disk blocks; they pass control of the tape in
129 * a ring via signals. The parent process traverses the filesystem and
130 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
131 * The following structure defines the instruction packets sent to slaves.
139 struct slave_results
{
140 ssize_t unclen
; /* uncompressed length */
141 ssize_t clen
; /* compressed length */
144 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
146 int tapea
; /* header number at start of this chunk */
147 int count
; /* count to next header (used for TS_TAPE */
149 int inode
; /* inode that we are currently dealing with */
150 int fd
; /* FD for this slave */
151 int pid
; /* PID for this slave */
152 int sent
; /* 1 == we've sent this slave requests */
153 int firstrec
; /* record number of this block */
154 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
155 struct req
*req
; /* buffer for requests */
159 char (*nextblock
)[TP_BSIZE
];
161 static time_t tstart_volume
; /* time of volume start */
162 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
164 int master
; /* pid of master, for sending error signals */
165 int tenths
; /* length of tape overhead per block written */
166 static int caught
; /* have we caught the signal to proceed? */
167 static int ready
; /* have we reached the lock point without having */
168 /* received the SIGUSR2 signal from the prev slave? */
169 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
170 /* SIGUSR2 arrives from the previous slave */
172 static int gtperr
= 0;
178 int pgoff
= getpagesize() - 1;
182 writesize
= ntrec
* TP_BSIZE
;
183 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
185 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
186 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
187 * repositioning after stopping, i.e, streaming mode, where the gap is
188 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
190 if (blocksperfile
== 0 && !unlimited
)
191 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
197 * Allocate tape buffer contiguous with the array of instruction
198 * packets, so flushtape() can write them together with one write().
199 * Align tape buffer on page boundary to speed up tape write().
201 for (i
= 0; i
<= SLAVES
; i
++) {
203 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
206 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
208 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
210 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
212 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
218 nextblock
= slp
->tblock
;
223 writerec(const void *dp
, int isspcl
)
226 slp
->req
[trecno
].dblk
= (daddr_t
)0;
227 slp
->req
[trecno
].count
= 1;
228 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
229 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
232 lastspclrec
= spcl
.c_tapea
;
240 dumpblock(daddr_t blkno
, int size
)
242 int avail
, tpblks
, dblkno
;
244 dblkno
= fsbtodb(sblock
, blkno
);
245 tpblks
= size
>> tp_bshift
;
246 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
247 slp
->req
[trecno
].dblk
= dblkno
;
248 slp
->req
[trecno
].count
= avail
;
250 spcl
.c_tapea
+= avail
;
253 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
265 msg("write error on %s: %s\n", tape
, strerror(errnum
));
266 quit("Cannot recover\n");
269 msg("write error %d blocks into volume %d: %s\n",
270 blocksthisvol
, tapeno
, strerror(errnum
));
271 broadcast("DUMP WRITE ERROR!\n");
272 if (query("Do you want to rewrite this volume?")) {
273 msg("Closing this volume. Prepare to restart with new media;\n");
274 msg("this dump volume will be rewritten.\n");
280 if (query("Do you want to start the next tape?"))
289 quit("Broken pipe\n");
294 * Update xferrate stats
303 ttaken
= tnow
- tstart_volume
;
304 blocks
= spcl
.c_tapea
- tapea_volume
;
305 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
307 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
308 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
310 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
311 long txfrate
= volkb
/ ttaken
;
312 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
313 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
314 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
318 double rate
= .0005 + (double) blocks
/ (double) volkb
;
319 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
321 tapeno
, blocks
, volkb
, rate
);
328 mktimeest(time_t tnow
)
330 static char msgbuf
[128];
335 if (blockswritten
< 500)
337 if (blockswritten
> tapesize
)
338 tapesize
= blockswritten
;
339 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
340 / blockswritten
* tapesize
;
341 if (tnow
> tstart_volume
)
342 (void)snprintf(msgbuf
, sizeof(msgbuf
),
343 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
344 (blockswritten
* 100.0) / tapesize
,
345 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
346 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
348 (void)snprintf(msgbuf
, sizeof(msgbuf
),
349 "%3.2f%% done, finished in %d:%02d\n",
350 (blockswritten
* 100.0) / tapesize
,
351 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
359 * information message upon receipt of SIGINFO
362 statussig(int notused
)
364 int save_errno
= errno
;
367 buf
= mktimeest(time(NULL
));
369 write(STDERR_FILENO
, buf
, strlen(buf
));
379 struct slave_results returned
;
381 int siz
= (char *)nextblock
- (char *)slp
->req
;
383 slp
->req
[trecno
].count
= 0; /* Sentinel */
385 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
386 quit("error writing command pipe: %s\n", strerror(errno
));
387 slp
->sent
= 1; /* we sent a request, read the response later */
389 lastfirstrec
= slp
->firstrec
;
391 if (++slp
>= &slaves
[SLAVES
])
394 /* Read results back from next slave */
396 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
397 != sizeof returned
) {
398 perror(" DUMP: error reading command pipe in master");
401 got
= returned
.unclen
;
402 bytes_written
+= returned
.clen
;
403 if (returned
.unclen
== returned
.clen
)
407 /* Check for errors or end of tape */
409 /* Check for errors */
413 msg("End of tape detected\n");
416 * Drain the results, don't care what the values were.
417 * If we read them here then trewind won't...
419 for (i
= 0; i
< SLAVES
; i
++) {
420 if (slaves
[i
].sent
) {
421 if (atomic_read( slaves
[i
].fd
,
422 (char *)&returned
, sizeof returned
)
423 != sizeof returned
) {
424 perror(" DUMP: error reading command pipe in master");
438 if (spcl
.c_type
!= TS_END
) {
439 for (i
= 0; i
< spcl
.c_count
; i
++)
440 if (spcl
.c_addr
[i
] != 0)
443 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
444 slp
->tapea
= spcl
.c_tapea
;
445 slp
->firstrec
= lastfirstrec
+ ntrec
;
447 nextblock
= slp
->tblock
;
449 asize
+= tenths
+ returned
.clen
/ density
;
450 blockswritten
+= ntrec
;
451 blocksthisvol
+= ntrec
;
452 if (!pipeout
&& !unlimited
) {
454 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= blocksperfile
* 1024
455 : blocksthisvol
>= blocksperfile
) {
460 else if (asize
> tsize
) {
473 struct slave_results returned
;
475 for (f
= 0; f
< SLAVES
; f
++) {
477 * Drain the results, but unlike EOT we DO (or should) care
478 * what the return values were, since if we detect EOT after
479 * we think we've written the last blocks to the tape anyway,
480 * we have to replay those blocks with rollforward.
482 * fixme: punt for now.
484 if (slaves
[f
].sent
) {
485 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
486 != sizeof returned
) {
487 perror(" DUMP: error reading command pipe in master");
490 got
= returned
.unclen
;
491 bytes_written
+= returned
.clen
;
492 if (returned
.unclen
== returned
.clen
)
500 msg("EOT detected in last 2 tape records!\n");
501 msg("Use a longer tape, decrease the size estimate\n");
502 quit("or use no size estimate at all.\n");
505 (void) close(slaves
[f
].fd
);
507 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
512 msg("Closing %s\n", tape
);
517 while (rmtopen(tape
, 0) < 0)
524 (void) close(tapefd
);
526 while ((f
= OPEN(tape
, 0)) < 0)
532 if (eot_script
&& spcl
.c_type
!= TS_END
) {
533 msg("Launching %s\n", eot_script
);
534 eot_code
= system_command(eot_script
, tape
, tapeno
);
536 if (eot_code
!= 0 && eot_code
!= 1) {
537 msg("Dump aborted by the end of tape script\n");
549 if (nexttape
|| Mflag
|| (eot_code
== 0) )
552 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
553 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
555 while (!query("Is the new volume mounted and ready to go?"))
556 if (query("Do you want to abort?")) {
565 register struct req
*p
, *q
, *prev
;
566 register struct slave
*tslp
;
567 int i
, size
, savedtapea
, got
;
568 union u_spcl
*ntb
, *otb
;
569 struct slave_results returned
;
574 tslp
= &slaves
[SLAVES
];
575 ntb
= (union u_spcl
*)tslp
->tblock
[1];
578 * Each of the N slaves should have requests that need to
579 * be replayed on the next tape. Use the extra slave buffers
580 * (slaves[SLAVES]) to construct request lists to be sent to
581 * each slave in turn.
583 for (i
= 0; i
< SLAVES
; i
++) {
585 otb
= (union u_spcl
*)slp
->tblock
;
588 * For each request in the current slave, copy it to tslp.
592 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
595 *ntb
++ = *otb
++; /* copy the datablock also */
600 quit("rollforward: protocol botch");
612 nextblock
= tslp
->tblock
;
613 savedtapea
= spcl
.c_tapea
;
614 spcl
.c_tapea
= slp
->tapea
;
616 spcl
.c_tapea
= savedtapea
;
617 lastspclrec
= savedtapea
- 1;
619 size
= (char *)ntb
- (char *)q
;
620 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
621 perror(" DUMP: error writing command pipe");
626 lastfirstrec
= slp
->firstrec
;
628 if (++slp
>= &slaves
[SLAVES
])
633 if (prev
->dblk
!= 0) {
635 * If the last one was a disk block, make the
636 * first of this one be the last bit of that disk
639 q
->dblk
= prev
->dblk
+
640 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
641 ntb
= (union u_spcl
*)tslp
->tblock
;
644 * It wasn't a disk block. Copy the data to its
645 * new location in the buffer.
648 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
649 ntb
= (union u_spcl
*)tslp
->tblock
[1];
653 nextblock
= slp
->tblock
;
656 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
657 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
664 * Clear the first slaves' response. One hopes that it
665 * worked ok, otherwise the tape is much too short!
668 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
669 != sizeof returned
) {
670 perror(" DUMP: error reading command pipe in master");
673 got
= returned
.unclen
;
674 bytes_written
+= returned
.clen
;
675 if (returned
.clen
== returned
.unclen
)
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
;
698 asize
+= tenths
+ returned
.clen
/ density
;
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
;
738 tapea_bytes
= bytes_written
;
739 tstart_volume
= time(NULL
);
743 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
745 (void)signal(SIGINT
, interrupt_save
);
748 * All signals are inherited...
752 msg("Context save fork fails in parent %d\n", parentpid
);
758 * save the context by waiting
759 * until the child doing all of the work returns.
760 * don't catch the interrupt
763 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
765 signal(SIGINT
, SIG_IGN
);
768 msg("Tape: %d; parent process: %d child process %d\n",
769 tapeno
+1, parentpid
, childpid
);
771 while ((waitpid
= wait(&status
)) != childpid
)
772 if (waitpid
!= rshpid
)
773 msg("Parent %d waiting for child %d has another child %d return\n",
774 parentpid
, childpid
, waitpid
);
776 msg("Child %d returns LOB status %o\n",
777 childpid
, status
&0xFF);
779 status
= (status
>> 8) & 0xFF;
783 msg("Child %d finishes X_FINOK\n", childpid
);
786 msg("Child %d finishes X_ABORT\n", childpid
);
789 msg("Child %d finishes X_REWRITE\n", childpid
);
792 msg("Child %d finishes unknown %d\n",
803 goto restore_check_point
;
805 msg("Bad return code from dump: %d\n", status
);
809 } else { /* we are the child; just continue */
811 sleep(4); /* allow time for parent's message to get out */
812 msg("Child on Tape %d has parent %d, my pid = %d\n",
813 tapeno
+1, parentpid
, getpid());
816 * If we have a name like "/dev/rmt0,/dev/rmt1",
817 * use the name before the comma first, and save
818 * the remaining names for subsequent volumes.
820 tapeno
++; /* current tape sequence */
822 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
823 tape
[MAXPATHLEN
- 1] = '\0';
824 msg("Dumping volume %d on %s\n", tapeno
, tape
);
826 else if (nexttape
|| strchr(tapeprefix
, ',')) {
827 if (nexttape
&& *nexttape
)
828 tapeprefix
= nexttape
;
829 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
834 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
835 tape
[MAXPATHLEN
- 1] = '\0';
836 msg("Dumping volume %d on %s\n", tapeno
, tape
);
839 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
841 OPEN(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
843 while ((tapefd
= (pipeout
? fileno(stdout
) :
844 OPEN(tape
, O_RDWR
|O_CREAT
, 0666))) < 0)
847 msg("Cannot open output \"%s\".\n", tape
);
848 if (!query("Do you want to retry the open?"))
852 enslave(); /* Share open tape file descriptor with slaves */
857 newtape
++; /* new tape signal */
858 spcl
.c_count
= slp
->count
;
860 * measure firstrec in TP_BSIZE units since restore doesn't
861 * know the correct ntrec value...
863 spcl
.c_firstrec
= slp
->firstrec
;
865 spcl
.c_type
= TS_TAPE
;
866 spcl
.c_flags
|= DR_NEWHEADER
;
867 spcl
.c_ntrec
= ntrec
;
869 spcl
.c_flags
|= DR_COMPRESSED
;
870 writeheader((dump_ino_t
)slp
->inode
);
871 spcl
.c_flags
&=~ DR_NEWHEADER
;
872 msg("Volume %d started with block %ld at: %s", tapeno
,
873 spcl
.c_tapea
, ctime(&tstart_volume
));
875 msg("Volume %d begins with blocks from inode %d\n",
884 if (master
!= 0 && master
!= getpid())
885 /* Signals master to call dumpabort */
886 (void) kill(master
, SIGTERM
);
889 msg("The ENTIRE dump is aborted.\n");
902 msg("pid = %d exits with status %d\n", getpid(), status
);
908 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
914 siglongjmp(jmpbuf
, 1);
922 #ifdef LINUX_FORK_BUG
930 { struct sigaction sa
;
931 memset(&sa
, 0, sizeof sa
);
932 sigemptyset(&sa
.sa_mask
);
933 sa
.sa_handler
= dumpabort
;
934 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
935 sa
.sa_handler
= sigpipe
;
936 sigaction(SIGPIPE
, &sa
, NULL
);
937 sa
.sa_handler
= proceed
;
938 sa
.sa_flags
= SA_RESTART
;
939 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
942 for (i
= 0; i
< SLAVES
; i
++) {
943 if (i
== slp
- &slaves
[0]) {
949 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
950 (slaves
[i
].pid
= fork()) < 0)
951 quit("too many slaves, %d (recompile smaller): %s\n",
954 slaves
[i
].fd
= cmd
[1];
956 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
958 for (j
= 0; j
<= i
; j
++)
959 (void) close(slaves
[j
].fd
);
961 sigaddset(&sigs
, SIGINT
); /* Master handles this */
963 sigaddset(&sigs
, SIGINFO
);
965 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
967 #ifdef LINUX_FORK_BUG
968 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
970 quit("master/slave protocol botched 3\n");
972 doslave(cmd
[0], i
, (slaves
[i
].pid
== slp
->pid
));
979 #ifdef LINUX_FORK_BUG
981 * Wait for all slaves to _actually_ start to circumvent a bug in
982 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
983 * returned from fork() causes a SEGV in the child process
985 for (i
= 0; i
< SLAVES
; i
++)
986 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
987 quit("master/slave protocol botched 4\n");
990 for (i
= 0; i
< SLAVES
; i
++)
991 (void) atomic_write( slaves
[i
].fd
,
992 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
993 sizeof slaves
[0].pid
);
1003 for (i
= 0; i
< SLAVES
; i
++)
1004 if (slaves
[i
].pid
> 0) {
1005 (void) kill(slaves
[i
].pid
, SIGKILL
);
1011 * Synchronization - each process waits for a SIGUSR2 from the
1012 * previous process before writing to the tape, and sends SIGUSR2
1013 * to the next process when the tape write completes. On tape errors
1014 * a SIGUSR1 is sent to the master which then terminates all of the
1018 doslave(int cmd
, int slave_number
, int first
)
1021 int nextslave
, size
, eot_count
, bufsize
;
1022 volatile int wrote
= 0;
1024 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1025 struct tapebuf
*comp_buf
= NULL
;
1026 int compresult
, do_compress
= !first
;
1027 unsigned long worklen
;
1029 unsigned int worklen2
;
1031 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1032 struct slave_results returns
;
1038 union u_spcl
*uspclptr
;
1039 struct s_spcl
*spclptr
;
1040 #endif /* USE_QFA */
1043 * Need our own seek pointer.
1045 (void) close(diskfd
);
1046 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1047 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1050 retval
= dump_fs_open(disk
, &fs
);
1052 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1053 #endif /* __linux__ */
1056 * Need the pid of the next slave in the loop...
1058 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1059 != sizeof nextslave
) {
1060 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1063 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1064 /* if we're doing a compressed dump, allocate the compress buffer */
1066 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1067 if (comp_buf
== NULL
)
1068 quit("couldn't allocate a compress buffer.\n");
1070 comp_buf
->flags
= COMPRESS_BZLIB
;
1072 comp_buf
->flags
= COMPRESS_ZLIB
;
1074 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1077 * Get list of blocks to dump, read the blocks into tape buffer
1079 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1080 register struct req
*p
= slp
->req
;
1082 for (trecno
= 0; trecno
< ntrec
;
1083 trecno
+= p
->count
, p
+= p
->count
) {
1084 if (p
->dblk
) { /* read a disk block */
1085 bread(p
->dblk
, slp
->tblock
[trecno
],
1086 p
->count
* TP_BSIZE
);
1087 } else { /* read record from pipe */
1088 if (p
->count
!= 1 || atomic_read( cmd
,
1089 (char *)slp
->tblock
[trecno
],
1090 TP_BSIZE
) != TP_BSIZE
)
1091 quit("master/slave protocol botched.\n");
1095 /* Try to write the data... */
1099 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1100 bufsize
= writesize
; /* length to write */
1101 returns
.clen
= returns
.unclen
= bufsize
;
1103 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1105 * When writing a compressed dump, each block except
1106 * the first one on each tape is written
1107 * from struct tapebuf with an 4 byte prefix
1108 * followed by the data. This can be less than
1109 * writesize. Restore, on a short read, can compare the
1110 * length read to the compressed length in the header
1111 * to verify that the read was good. Blocks which don't
1112 * compress well are written uncompressed.
1113 * The first block written by each slave is not compressed
1114 * and does not have a prefix.
1117 if (compressed
&& do_compress
) {
1118 comp_buf
->length
= bufsize
;
1119 worklen
= TP_BSIZE
+ writesize
;
1122 compresult
= compress2(comp_buf
->buf
,
1124 (char *)slp
->tblock
[0],
1127 if (compresult
== Z_OK
)
1132 #endif /* HAVE_ZLIB */
1136 compresult
= BZ2_bzBuffToBuffCompress(
1139 (char *)slp
->tblock
[0],
1144 if (compresult
== BZ_OK
)
1150 #endif /* HAVE_BZLIB */
1151 if (compresult
&& worklen
<= (writesize
- 16)) {
1152 /* write the compressed buffer */
1153 comp_buf
->length
= worklen
;
1154 comp_buf
->compressed
= 1;
1155 buffer
= (char *) comp_buf
;
1156 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1159 /* write the data uncompressed */
1160 comp_buf
->length
= writesize
;
1161 comp_buf
->compressed
= 0;
1162 buffer
= (char *) comp_buf
;
1163 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1164 returns
.unclen
= returns
.clen
;
1165 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1168 /* compress the remaining blocks if we're compressing */
1169 do_compress
= compressed
;
1170 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1172 if (sigsetjmp(jmpbuf
, 1) == 0) {
1181 if (gTapeposfd
>= 0) {
1182 uspclptr
= (union u_spcl
*)&slp
->tblock
[0];
1183 spclptr
= &uspclptr
->s_spcl
;
1184 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1185 (spclptr
->c_type
== TS_INODE
) &&
1186 (spclptr
->c_date
== gThisDumpDate
)) {
1187 /* if an error occured previously don't
1190 if ((gtperr
= GetTapePos(&curtapepos
)) == 0) {
1192 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1194 sprintf(gTps
, "%ld\t%d\t%ld\n", (unsigned long)spclptr
->c_inumber
, tapeno
, curtapepos
);
1195 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1196 warn("error writing tapepos file.\n");
1202 #endif /* USE_QFA */
1204 while (eot_count
< 10 && size
< bufsize
) {
1207 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1210 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1212 printf("slave %d wrote %d\n", slave_number
, wrote
);
1222 if (size
!= bufsize
)
1223 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1224 slave_number
, size
, bufsize
);
1228 * Handle ENOSPC as an EOT condition.
1230 if (wrote
< 0 && errno
== ENOSPC
) {
1236 returns
.clen
= returns
.unclen
= 0;
1239 * pass errno back to master for special handling
1242 returns
.unclen
= -errno
;
1245 * pass size of data and size of write back to master
1246 * (for EOT handling)
1248 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1251 * Signal the next slave to go.
1253 (void) kill(nextslave
, SIGUSR2
);
1256 quit("error reading command pipe: %s\n", strerror(errno
));
1260 * Since a read from a pipe may not return all we asked for,
1261 * or a write may not write all we ask if we get a signal,
1262 * loop until the count is satisfied (or error).
1265 atomic_read(int fd
, void *buf
, size_t count
)
1267 int got
, need
= count
;
1270 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1272 } while (got
== -1 && errno
== EINTR
);
1273 return (got
< 0 ? got
: count
- need
);
1277 * Since a read from a pipe may not return all we asked for,
1278 * or a write may not write all we ask if we get a signal,
1279 * loop until the count is satisfied (or error).
1282 atomic_write(int fd
, const void *buf
, size_t count
)
1284 int got
, need
= count
;
1287 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1289 } while (got
== -1 && errno
== EINTR
);
1290 return (got
< 0 ? got
: count
- need
);
1296 * read the current tape position
1299 GetTapePos(long *pos
)
1304 if (ioctl(tapefd
, MTIOCPOS
, pos
) == -1) {
1306 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1312 #endif /* USE_QFA */