]>
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 1.57 2001/09/12 10:21:49 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 #undef atomic_read /* this get wrongly defined in kernel */
76 /* headers and we don't want it */
77 #ifdef HAVE_EXT2FS_EXT2_FS_H
78 #include <ext2fs/ext2_fs.h>
80 #include <linux/ext2_fs.h>
82 #include <ext2fs/ext2fs.h>
83 #include <bsdcompat.h>
85 #include <sys/vnode.h>
88 #include <ufs/inode.h>
90 #include <ufs/ufs/dinode.h>
91 #include <ufs/ffs/fs.h>
92 #endif /* __linux__ */
94 #include <protocols/dumprestore.h>
98 #endif /* HAVE_ZLIB */
102 #endif /* HAVE_BZLIB */
106 int writesize
; /* size of malloc()ed buffer for tape */
107 long lastspclrec
= -1; /* tape block number of last written header */
108 int trecno
= 0; /* next record to write in current block */
109 extern long blocksperfile
; /* number of blocks per output file */
110 long blocksthisvol
; /* number of blocks on current output file */
111 extern int ntrec
; /* blocking factor on tape */
112 extern int cartridge
;
117 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
119 static ssize_t atomic_read
__P((int, void *, size_t));
120 static ssize_t atomic_write
__P((int, const void *, size_t));
121 static void doslave
__P((int, int, int));
122 static void enslave
__P((void));
123 static void flushtape
__P((void));
124 static void killall
__P((void));
125 static void rollforward
__P((void));
128 * Concurrent dump mods (Caltech) - disk block reading and tape writing
129 * are exported to several slave processes. While one slave writes the
130 * tape, the others read disk blocks; they pass control of the tape in
131 * a ring via signals. The parent process traverses the filesystem and
132 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
133 * The following structure defines the instruction packets sent to slaves.
141 struct slave_results
{
142 ssize_t unclen
; /* uncompressed length */
143 ssize_t clen
; /* compressed length */
146 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
148 int tapea
; /* header number at start of this chunk */
149 int count
; /* count to next header (used for TS_TAPE */
151 int inode
; /* inode that we are currently dealing with */
152 int fd
; /* FD for this slave */
153 int pid
; /* PID for this slave */
154 int sent
; /* 1 == we've sent this slave requests */
155 int firstrec
; /* record number of this block */
156 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
157 struct req
*req
; /* buffer for requests */
161 char (*nextblock
)[TP_BSIZE
];
163 static time_t tstart_volume
; /* time of volume start */
164 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
166 int master
; /* pid of master, for sending error signals */
167 int tenths
; /* length of tape overhead per block written */
168 static int caught
; /* have we caught the signal to proceed? */
169 static int ready
; /* have we reached the lock point without having */
170 /* received the SIGUSR2 signal from the prev slave? */
171 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
172 /* SIGUSR2 arrives from the previous slave */
174 static int gtperr
= 0;
180 int pgoff
= getpagesize() - 1;
184 writesize
= ntrec
* TP_BSIZE
;
185 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
187 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
188 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
189 * repositioning after stopping, i.e, streaming mode, where the gap is
190 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
192 if (blocksperfile
== 0 && !unlimited
)
193 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
199 * Allocate tape buffer contiguous with the array of instruction
200 * packets, so flushtape() can write them together with one write().
201 * Align tape buffer on page boundary to speed up tape write().
203 for (i
= 0; i
<= SLAVES
; i
++) {
205 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
208 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
210 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
212 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
214 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
220 nextblock
= slp
->tblock
;
225 writerec(const void *dp
, int isspcl
)
228 slp
->req
[trecno
].dblk
= (daddr_t
)0;
229 slp
->req
[trecno
].count
= 1;
230 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
231 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
234 lastspclrec
= spcl
.c_tapea
;
242 dumpblock(daddr_t blkno
, int size
)
244 int avail
, tpblks
, dblkno
;
246 dblkno
= fsbtodb(sblock
, blkno
);
247 tpblks
= size
>> tp_bshift
;
248 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
249 slp
->req
[trecno
].dblk
= dblkno
;
250 slp
->req
[trecno
].count
= avail
;
252 spcl
.c_tapea
+= avail
;
255 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
267 msg("write error on %s: %s\n", tape
, strerror(errnum
));
268 quit("Cannot recover\n");
271 msg("write error %d blocks into volume %d: %s\n",
272 blocksthisvol
, tapeno
, strerror(errnum
));
273 broadcast("DUMP WRITE ERROR!\n");
274 if (query("Do you want to rewrite this volume?")) {
275 msg("Closing this volume. Prepare to restart with new media;\n");
276 msg("this dump volume will be rewritten.\n");
282 if (query("Do you want to start the next tape?"))
291 quit("Broken pipe\n");
296 * Update xferrate stats
305 ttaken
= tnow
- tstart_volume
;
306 blocks
= spcl
.c_tapea
- tapea_volume
;
307 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
309 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
310 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
312 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
313 long txfrate
= volkb
/ ttaken
;
314 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
315 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
316 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
320 double rate
= .0005 + (double) blocks
/ (double) volkb
;
321 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
323 tapeno
, blocks
, volkb
, rate
);
330 mktimeest(time_t tnow
)
332 static char msgbuf
[128];
337 if (blockswritten
< 500)
339 if (blockswritten
> tapesize
)
340 tapesize
= blockswritten
;
341 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
342 / blockswritten
* tapesize
;
343 if (tnow
> tstart_volume
)
344 (void)snprintf(msgbuf
, sizeof(msgbuf
),
345 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
346 (blockswritten
* 100.0) / tapesize
,
347 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
348 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
350 (void)snprintf(msgbuf
, sizeof(msgbuf
),
351 "%3.2f%% done, finished in %d:%02d\n",
352 (blockswritten
* 100.0) / tapesize
,
353 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
361 * information message upon receipt of SIGINFO
364 statussig(int notused
)
366 int save_errno
= errno
;
369 buf
= mktimeest(time(NULL
));
371 write(STDERR_FILENO
, buf
, strlen(buf
));
381 struct slave_results returned
;
383 int siz
= (char *)nextblock
- (char *)slp
->req
;
385 slp
->req
[trecno
].count
= 0; /* Sentinel */
387 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
388 quit("error writing command pipe: %s\n", strerror(errno
));
389 slp
->sent
= 1; /* we sent a request, read the response later */
391 lastfirstrec
= slp
->firstrec
;
393 if (++slp
>= &slaves
[SLAVES
])
396 /* Read results back from next slave */
398 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
399 != sizeof returned
) {
400 perror(" DUMP: error reading command pipe in master");
403 got
= returned
.unclen
;
404 bytes_written
+= returned
.clen
;
405 if (returned
.unclen
== returned
.clen
)
409 /* Check for errors or end of tape */
411 /* Check for errors */
415 msg("End of tape detected\n");
418 * Drain the results, don't care what the values were.
419 * If we read them here then trewind won't...
421 for (i
= 0; i
< SLAVES
; i
++) {
422 if (slaves
[i
].sent
) {
423 if (atomic_read( slaves
[i
].fd
,
424 (char *)&returned
, sizeof returned
)
425 != sizeof returned
) {
426 perror(" DUMP: error reading command pipe in master");
440 if (spcl
.c_type
!= TS_END
) {
441 for (i
= 0; i
< spcl
.c_count
; i
++)
442 if (spcl
.c_addr
[i
] != 0)
445 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
446 slp
->tapea
= spcl
.c_tapea
;
447 slp
->firstrec
= lastfirstrec
+ ntrec
;
449 nextblock
= slp
->tblock
;
451 asize
+= tenths
+ returned
.clen
/ density
;
452 blockswritten
+= ntrec
;
453 blocksthisvol
+= ntrec
;
454 if (!pipeout
&& !unlimited
) {
456 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= blocksperfile
* 1024
457 : blocksthisvol
>= blocksperfile
) {
462 else if (asize
> tsize
) {
475 struct slave_results returned
;
477 for (f
= 0; f
< SLAVES
; f
++) {
479 * Drain the results, but unlike EOT we DO (or should) care
480 * what the return values were, since if we detect EOT after
481 * we think we've written the last blocks to the tape anyway,
482 * we have to replay those blocks with rollforward.
484 * fixme: punt for now.
486 if (slaves
[f
].sent
) {
487 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
488 != sizeof returned
) {
489 perror(" DUMP: error reading command pipe in master");
492 got
= returned
.unclen
;
493 bytes_written
+= returned
.clen
;
494 if (returned
.unclen
== returned
.clen
)
502 msg("EOT detected in last 2 tape records!\n");
503 msg("Use a longer tape, decrease the size estimate\n");
504 quit("or use no size estimate at all.\n");
507 (void) close(slaves
[f
].fd
);
509 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
514 msg("Closing %s\n", tape
);
519 while (rmtopen(tape
, 0) < 0)
526 (void) close(tapefd
);
528 while ((f
= OPEN(tape
, 0)) < 0)
534 if (eot_script
&& spcl
.c_type
!= TS_END
) {
535 msg("Launching %s\n", eot_script
);
536 eot_code
= system_command(eot_script
, tape
, tapeno
);
538 if (eot_code
!= 0 && eot_code
!= 1) {
539 msg("Dump aborted by the end of tape script\n");
551 if (nexttape
|| Mflag
|| (eot_code
== 0) )
554 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
555 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
557 while (!query("Is the new volume mounted and ready to go?"))
558 if (query("Do you want to abort?")) {
567 register struct req
*p
, *q
= NULL
, *prev
;
568 register struct slave
*tslp
;
569 int i
, size
, savedtapea
, got
;
570 union u_spcl
*ntb
, *otb
;
571 struct slave_results returned
;
576 tslp
= &slaves
[SLAVES
];
577 ntb
= (union u_spcl
*)tslp
->tblock
[1];
580 * Each of the N slaves should have requests that need to
581 * be replayed on the next tape. Use the extra slave buffers
582 * (slaves[SLAVES]) to construct request lists to be sent to
583 * each slave in turn.
585 for (i
= 0; i
< SLAVES
; i
++) {
587 otb
= (union u_spcl
*)slp
->tblock
;
590 * For each request in the current slave, copy it to tslp.
594 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
597 *ntb
++ = *otb
++; /* copy the datablock also */
602 quit("rollforward: protocol botch");
614 nextblock
= tslp
->tblock
;
615 savedtapea
= spcl
.c_tapea
;
616 spcl
.c_tapea
= slp
->tapea
;
618 spcl
.c_tapea
= savedtapea
;
619 lastspclrec
= savedtapea
- 1;
621 size
= (char *)ntb
- (char *)q
;
622 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
623 perror(" DUMP: error writing command pipe");
628 lastfirstrec
= slp
->firstrec
;
630 if (++slp
>= &slaves
[SLAVES
])
635 if (prev
->dblk
!= 0) {
637 * If the last one was a disk block, make the
638 * first of this one be the last bit of that disk
641 q
->dblk
= prev
->dblk
+
642 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
643 ntb
= (union u_spcl
*)tslp
->tblock
;
646 * It wasn't a disk block. Copy the data to its
647 * new location in the buffer.
650 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
651 ntb
= (union u_spcl
*)tslp
->tblock
[1];
655 nextblock
= slp
->tblock
;
658 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
659 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
666 * Clear the first slaves' response. One hopes that it
667 * worked ok, otherwise the tape is much too short!
670 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
671 != sizeof returned
) {
672 perror(" DUMP: error reading command pipe in master");
675 got
= returned
.unclen
;
676 bytes_written
+= returned
.clen
;
677 if (returned
.clen
== returned
.unclen
)
685 quit("EOT detected at start of the tape!\n");
691 if (spcl
.c_type
!= TS_END
) {
692 for (i
= 0; i
< spcl
.c_count
; i
++)
693 if (spcl
.c_addr
[i
] != 0)
697 slp
->firstrec
= lastfirstrec
+ ntrec
;
698 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
700 asize
+= tenths
+ returned
.clen
/ density
;
701 blockswritten
+= ntrec
;
702 blocksthisvol
+= ntrec
;
707 * We implement taking and restoring checkpoints on the tape level.
708 * When each tape is opened, a new process is created by forking; this
709 * saves all of the necessary context in the parent. The child
710 * continues the dump; the parent waits around, saving the context.
711 * If the child returns X_REWRITE, then it had problems writing that tape;
712 * this causes the parent to fork again, duplicating the context, and
713 * everything continues as if nothing had happened.
716 startnewtape(int top
)
727 sigaddset(&sigs
, SIGINT
);
728 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
729 #else /* __linux__ */
731 void (*interrupt_save
)();
733 sig_t interrupt_save
;
735 interrupt_save
= signal(SIGINT
, SIG_IGN
);
736 #endif /* __linux__ */
738 parentpid
= getpid();
739 tapea_volume
= spcl
.c_tapea
;
740 tapea_bytes
= bytes_written
;
741 tstart_volume
= time(NULL
);
745 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
747 (void)signal(SIGINT
, interrupt_save
);
750 * All signals are inherited...
754 msg("Context save fork fails in parent %d\n", parentpid
);
760 * save the context by waiting
761 * until the child doing all of the work returns.
762 * don't catch the interrupt
765 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
767 signal(SIGINT
, SIG_IGN
);
770 msg("Tape: %d; parent process: %d child process %d\n",
771 tapeno
+1, parentpid
, childpid
);
773 while ((waitpid
= wait(&status
)) != childpid
)
774 if (waitpid
!= rshpid
)
775 msg("Parent %d waiting for child %d has another child %d return\n",
776 parentpid
, childpid
, waitpid
);
778 msg("Child %d returns LOB status %o\n",
779 childpid
, status
&0xFF);
781 status
= (status
>> 8) & 0xFF;
785 msg("Child %d finishes X_FINOK\n", childpid
);
788 msg("Child %d finishes X_ABORT\n", childpid
);
791 msg("Child %d finishes X_REWRITE\n", childpid
);
794 msg("Child %d finishes unknown %d\n",
805 goto restore_check_point
;
807 msg("Bad return code from dump: %d\n", status
);
811 } else { /* we are the child; just continue */
813 sleep(4); /* allow time for parent's message to get out */
814 msg("Child on Tape %d has parent %d, my pid = %d\n",
815 tapeno
+1, parentpid
, getpid());
818 * If we have a name like "/dev/rmt0,/dev/rmt1",
819 * use the name before the comma first, and save
820 * the remaining names for subsequent volumes.
822 tapeno
++; /* current tape sequence */
824 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
825 tape
[MAXPATHLEN
- 1] = '\0';
826 msg("Dumping volume %d on %s\n", tapeno
, tape
);
828 else if (nexttape
|| strchr(tapeprefix
, ',')) {
829 if (nexttape
&& *nexttape
)
830 tapeprefix
= nexttape
;
831 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
836 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
837 tape
[MAXPATHLEN
- 1] = '\0';
838 msg("Dumping volume %d on %s\n", tapeno
, tape
);
841 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
843 OPEN(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
845 while ((tapefd
= (pipeout
? fileno(stdout
) :
846 OPEN(tape
, O_RDWR
|O_CREAT
, 0666))) < 0)
849 msg("Cannot open output \"%s\".\n", tape
);
850 if (!query("Do you want to retry the open?"))
854 enslave(); /* Share open tape file descriptor with slaves */
859 newtape
++; /* new tape signal */
860 spcl
.c_count
= slp
->count
;
862 * measure firstrec in TP_BSIZE units since restore doesn't
863 * know the correct ntrec value...
865 spcl
.c_firstrec
= slp
->firstrec
;
867 spcl
.c_type
= TS_TAPE
;
868 spcl
.c_flags
|= DR_NEWHEADER
;
869 spcl
.c_ntrec
= ntrec
;
871 spcl
.c_flags
|= DR_COMPRESSED
;
872 writeheader((dump_ino_t
)slp
->inode
);
873 spcl
.c_flags
&=~ DR_NEWHEADER
;
874 msg("Volume %d started with block %ld at: %s", tapeno
,
875 spcl
.c_tapea
, ctime(&tstart_volume
));
877 msg("Volume %d begins with blocks from inode %d\n",
886 if (master
!= 0 && master
!= getpid())
887 /* Signals master to call dumpabort */
888 (void) kill(master
, SIGTERM
);
891 msg("The ENTIRE dump is aborted.\n");
904 msg("pid = %d exits with status %d\n", getpid(), status
);
910 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
916 siglongjmp(jmpbuf
, 1);
924 #ifdef LINUX_FORK_BUG
932 { struct sigaction sa
;
933 memset(&sa
, 0, sizeof sa
);
934 sigemptyset(&sa
.sa_mask
);
935 sa
.sa_handler
= dumpabort
;
936 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
937 sa
.sa_handler
= sigpipe
;
938 sigaction(SIGPIPE
, &sa
, NULL
);
939 sa
.sa_handler
= proceed
;
940 sa
.sa_flags
= SA_RESTART
;
941 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
944 for (i
= 0; i
< SLAVES
; i
++) {
945 if (i
== slp
- &slaves
[0]) {
951 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
952 (slaves
[i
].pid
= fork()) < 0)
953 quit("too many slaves, %d (recompile smaller): %s\n",
956 slaves
[i
].fd
= cmd
[1];
958 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
960 for (j
= 0; j
<= i
; j
++)
961 (void) close(slaves
[j
].fd
);
963 sigaddset(&sigs
, SIGINT
); /* Master handles this */
965 sigaddset(&sigs
, SIGINFO
);
967 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
969 #ifdef LINUX_FORK_BUG
970 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
972 quit("master/slave protocol botched 3\n");
974 doslave(cmd
[0], i
, (slaves
[i
].pid
== slp
->pid
));
981 #ifdef LINUX_FORK_BUG
983 * Wait for all slaves to _actually_ start to circumvent a bug in
984 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
985 * returned from fork() causes a SEGV in the child process
987 for (i
= 0; i
< SLAVES
; i
++)
988 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
989 quit("master/slave protocol botched 4\n");
992 for (i
= 0; i
< SLAVES
; i
++)
993 (void) atomic_write( slaves
[i
].fd
,
994 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
995 sizeof slaves
[0].pid
);
1005 for (i
= 0; i
< SLAVES
; i
++)
1006 if (slaves
[i
].pid
> 0) {
1007 (void) kill(slaves
[i
].pid
, SIGKILL
);
1013 * Synchronization - each process waits for a SIGUSR2 from the
1014 * previous process before writing to the tape, and sends SIGUSR2
1015 * to the next process when the tape write completes. On tape errors
1016 * a SIGUSR1 is sent to the master which then terminates all of the
1020 doslave(int cmd
, int slave_number
, int first
)
1023 int nextslave
, size
, eot_count
, bufsize
;
1024 volatile int wrote
= 0;
1026 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1027 struct tapebuf
*comp_buf
= NULL
;
1028 int compresult
, do_compress
= !first
;
1029 unsigned long worklen
;
1031 unsigned int worklen2
;
1033 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1034 struct slave_results returns
;
1040 union u_spcl
*uspclptr
;
1041 struct s_spcl
*spclptr
;
1042 #endif /* USE_QFA */
1045 * Need our own seek pointer.
1047 (void) close(diskfd
);
1048 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1049 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1052 (void)ioctl(diskfd
, BLKFLSBUF
);
1055 retval
= dump_fs_open(disk
, &fs
);
1057 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1058 #endif /* __linux__ */
1061 * Need the pid of the next slave in the loop...
1063 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1064 != sizeof nextslave
) {
1065 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1068 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1069 /* if we're doing a compressed dump, allocate the compress buffer */
1071 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1072 if (comp_buf
== NULL
)
1073 quit("couldn't allocate a compress buffer.\n");
1075 comp_buf
->flags
= COMPRESS_BZLIB
;
1077 comp_buf
->flags
= COMPRESS_ZLIB
;
1079 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1082 * Get list of blocks to dump, read the blocks into tape buffer
1084 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1085 register struct req
*p
= slp
->req
;
1087 for (trecno
= 0; trecno
< ntrec
;
1088 trecno
+= p
->count
, p
+= p
->count
) {
1089 if (p
->dblk
) { /* read a disk block */
1090 bread(p
->dblk
, slp
->tblock
[trecno
],
1091 p
->count
* TP_BSIZE
);
1092 } else { /* read record from pipe */
1093 if (p
->count
!= 1 || atomic_read( cmd
,
1094 (char *)slp
->tblock
[trecno
],
1095 TP_BSIZE
) != TP_BSIZE
)
1096 quit("master/slave protocol botched.\n");
1100 /* Try to write the data... */
1104 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1105 bufsize
= writesize
; /* length to write */
1106 returns
.clen
= returns
.unclen
= bufsize
;
1108 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1110 * When writing a compressed dump, each block except
1111 * the first one on each tape is written
1112 * from struct tapebuf with an 4 byte prefix
1113 * followed by the data. This can be less than
1114 * writesize. Restore, on a short read, can compare the
1115 * length read to the compressed length in the header
1116 * to verify that the read was good. Blocks which don't
1117 * compress well are written uncompressed.
1118 * The first block written by each slave is not compressed
1119 * and does not have a prefix.
1122 if (compressed
&& do_compress
) {
1123 comp_buf
->length
= bufsize
;
1124 worklen
= TP_BSIZE
+ writesize
;
1128 compresult
= compress2(comp_buf
->buf
,
1130 (char *)slp
->tblock
[0],
1133 if (compresult
== Z_OK
)
1138 #endif /* HAVE_ZLIB */
1142 compresult
= BZ2_bzBuffToBuffCompress(
1145 (char *)slp
->tblock
[0],
1150 if (compresult
== BZ_OK
)
1156 #endif /* HAVE_BZLIB */
1157 if (compresult
&& worklen
<= (writesize
- 16)) {
1158 /* write the compressed buffer */
1159 comp_buf
->length
= worklen
;
1160 comp_buf
->compressed
= 1;
1161 buffer
= (char *) comp_buf
;
1162 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1165 /* write the data uncompressed */
1166 comp_buf
->length
= writesize
;
1167 comp_buf
->compressed
= 0;
1168 buffer
= (char *) comp_buf
;
1169 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1170 returns
.unclen
= returns
.clen
;
1171 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1174 /* compress the remaining blocks if we're compressing */
1175 do_compress
= compressed
;
1176 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1178 if (sigsetjmp(jmpbuf
, 1) == 0) {
1187 if (gTapeposfd
>= 0) {
1188 uspclptr
= (union u_spcl
*)&slp
->tblock
[0];
1189 spclptr
= &uspclptr
->s_spcl
;
1190 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1191 (spclptr
->c_type
== TS_INODE
) &&
1192 (spclptr
->c_date
== gThisDumpDate
)) {
1193 /* if an error occured previously don't
1196 if ((gtperr
= GetTapePos(&curtapepos
)) == 0) {
1198 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1200 sprintf(gTps
, "%ld\t%d\t%ld\n", (unsigned long)spclptr
->c_inumber
, tapeno
, curtapepos
);
1201 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1202 warn("error writing tapepos file.\n");
1208 #endif /* USE_QFA */
1210 while (eot_count
< 10 && size
< bufsize
) {
1213 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1216 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1218 printf("slave %d wrote %d\n", slave_number
, wrote
);
1228 if (size
!= bufsize
)
1229 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1230 slave_number
, size
, bufsize
);
1234 * Handle ENOSPC as an EOT condition.
1236 if (wrote
< 0 && errno
== ENOSPC
) {
1242 returns
.clen
= returns
.unclen
= 0;
1245 * pass errno back to master for special handling
1248 returns
.unclen
= -errno
;
1251 * pass size of data and size of write back to master
1252 * (for EOT handling)
1254 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1257 * Signal the next slave to go.
1259 (void) kill(nextslave
, SIGUSR2
);
1262 quit("error reading command pipe: %s\n", strerror(errno
));
1266 * Since a read from a pipe may not return all we asked for,
1267 * or a write may not write all we ask if we get a signal,
1268 * loop until the count is satisfied (or error).
1271 atomic_read(int fd
, void *buf
, size_t count
)
1273 int got
, need
= count
;
1276 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1278 } while (got
== -1 && errno
== EINTR
);
1279 return (got
< 0 ? got
: count
- need
);
1283 * Since a read from a pipe may not return all we asked for,
1284 * or a write may not write all we ask if we get a signal,
1285 * loop until the count is satisfied (or error).
1288 atomic_write(int fd
, const void *buf
, size_t count
)
1290 int got
, need
= count
;
1293 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1295 } while (got
== -1 && errno
== EINTR
);
1296 return (got
< 0 ? got
: count
- need
);
1302 * read the current tape position
1305 GetTapePos(long *pos
)
1310 if (ioctl(tapefd
, MTIOCPOS
, pos
) == -1) {
1312 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1318 #endif /* USE_QFA */