]>
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 <stelian@popies.net>, 1999-2000
6 * Stelian Pop <stelian@popies.net> - AlcĂ´ve <www.alcove.com>, 2000-2002
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 1.60 2002/01/22 11:12:28 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 */
118 static int magtapeout
; /* output is really a tape */
120 static ssize_t atomic_read
__P((int, void *, size_t));
121 static ssize_t atomic_write
__P((int, const void *, size_t));
122 static void doslave
__P((int, int, int));
123 static void enslave
__P((void));
124 static void flushtape
__P((void));
125 static void killall
__P((void));
126 static void rollforward
__P((void));
128 static int GetTapePos
__P((long *));
129 static void MkTapeString
__P((struct s_spcl
*, long));
133 * Concurrent dump mods (Caltech) - disk block reading and tape writing
134 * are exported to several slave processes. While one slave writes the
135 * tape, the others read disk blocks; they pass control of the tape in
136 * a ring via signals. The parent process traverses the filesystem and
137 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
138 * The following structure defines the instruction packets sent to slaves.
146 struct slave_results
{
147 ssize_t unclen
; /* uncompressed length */
148 ssize_t clen
; /* compressed length */
151 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
153 int tapea
; /* header number at start of this chunk */
154 int count
; /* count to next header (used for TS_TAPE */
156 int inode
; /* inode that we are currently dealing with */
157 int fd
; /* FD for this slave */
158 int pid
; /* PID for this slave */
159 int sent
; /* 1 == we've sent this slave requests */
160 int firstrec
; /* record number of this block */
161 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
162 struct req
*req
; /* buffer for requests */
166 char (*nextblock
)[TP_BSIZE
];
168 static time_t tstart_volume
; /* time of volume start */
169 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
171 int master
; /* pid of master, for sending error signals */
172 int tenths
; /* length of tape overhead per block written */
173 static int caught
; /* have we caught the signal to proceed? */
174 static int ready
; /* have we reached the lock point without having */
175 /* received the SIGUSR2 signal from the prev slave? */
176 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
177 /* SIGUSR2 arrives from the previous slave */
179 static int gtperr
= 0;
185 int pgoff
= getpagesize() - 1;
189 writesize
= ntrec
* TP_BSIZE
;
190 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
192 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
193 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
194 * repositioning after stopping, i.e, streaming mode, where the gap is
195 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
197 if (blocksperfile
== 0 && !unlimited
)
198 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
204 * Allocate tape buffer contiguous with the array of instruction
205 * packets, so flushtape() can write them together with one write().
206 * Align tape buffer on page boundary to speed up tape write().
208 for (i
= 0; i
<= SLAVES
; i
++) {
210 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
213 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
215 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
217 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
219 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
225 nextblock
= slp
->tblock
;
230 writerec(const void *dp
, int isspcl
)
233 slp
->req
[trecno
].dblk
= (daddr_t
)0;
234 slp
->req
[trecno
].count
= 1;
235 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
236 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
239 lastspclrec
= spcl
.c_tapea
;
247 dumpblock(daddr_t blkno
, int size
)
249 int avail
, tpblks
, dblkno
;
251 dblkno
= fsbtodb(sblock
, blkno
);
252 tpblks
= size
>> tp_bshift
;
253 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
254 slp
->req
[trecno
].dblk
= dblkno
;
255 slp
->req
[trecno
].count
= avail
;
257 spcl
.c_tapea
+= avail
;
260 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
272 msg("write error on %s: %s\n", tape
, strerror(errnum
));
273 quit("Cannot recover\n");
276 msg("write error %d blocks into volume %d: %s\n",
277 blocksthisvol
, tapeno
, strerror(errnum
));
278 broadcast("DUMP WRITE ERROR!\n");
279 if (query("Do you want to rewrite this volume?")) {
280 msg("Closing this volume. Prepare to restart with new media;\n");
281 msg("this dump volume will be rewritten.\n");
287 if (query("Do you want to start the next tape?"))
296 quit("Broken pipe\n");
301 * Update xferrate stats
310 ttaken
= tnow
- tstart_volume
;
311 blocks
= spcl
.c_tapea
- tapea_volume
;
312 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
314 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
315 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
317 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
318 long txfrate
= volkb
/ ttaken
;
319 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
320 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
321 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
325 double rate
= .0005 + (double) blocks
/ (double) volkb
;
326 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
328 tapeno
, blocks
, volkb
, rate
);
335 mktimeest(time_t tnow
)
337 static char msgbuf
[128];
342 if (blockswritten
< 500)
344 if (blockswritten
> tapesize
)
345 tapesize
= blockswritten
;
346 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
347 / blockswritten
* tapesize
;
348 if (tnow
> tstart_volume
)
349 (void)snprintf(msgbuf
, sizeof(msgbuf
),
350 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
351 (blockswritten
* 100.0) / tapesize
,
352 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
353 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
355 (void)snprintf(msgbuf
, sizeof(msgbuf
),
356 "%3.2f%% done, finished in %d:%02d\n",
357 (blockswritten
* 100.0) / tapesize
,
358 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
366 * information message upon receipt of SIGINFO
369 statussig(int notused
)
371 int save_errno
= errno
;
374 buf
= mktimeest(time(NULL
));
376 write(STDERR_FILENO
, buf
, strlen(buf
));
386 struct slave_results returned
;
388 int siz
= (char *)nextblock
- (char *)slp
->req
;
390 slp
->req
[trecno
].count
= 0; /* Sentinel */
392 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
393 quit("error writing command pipe: %s\n", strerror(errno
));
394 slp
->sent
= 1; /* we sent a request, read the response later */
396 lastfirstrec
= slp
->firstrec
;
398 if (++slp
>= &slaves
[SLAVES
])
401 /* Read results back from next slave */
403 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
404 != sizeof returned
) {
405 perror(" DUMP: error reading command pipe in master");
408 got
= returned
.unclen
;
409 bytes_written
+= returned
.clen
;
410 if (returned
.unclen
== returned
.clen
)
414 /* Check for errors or end of tape */
416 /* Check for errors */
420 msg("End of tape detected\n");
423 * Drain the results, don't care what the values were.
424 * If we read them here then trewind won't...
426 for (i
= 0; i
< SLAVES
; i
++) {
427 if (slaves
[i
].sent
) {
428 if (atomic_read( slaves
[i
].fd
,
429 (char *)&returned
, sizeof returned
)
430 != sizeof returned
) {
431 perror(" DUMP: error reading command pipe in master");
445 if (spcl
.c_type
!= TS_END
) {
446 for (i
= 0; i
< spcl
.c_count
; i
++)
447 if (spcl
.c_addr
[i
] != 0)
450 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
451 slp
->tapea
= spcl
.c_tapea
;
452 slp
->firstrec
= lastfirstrec
+ ntrec
;
454 nextblock
= slp
->tblock
;
456 asize
+= tenths
+ returned
.clen
/ density
;
457 blockswritten
+= ntrec
;
458 blocksthisvol
+= ntrec
;
459 if (!pipeout
&& !unlimited
) {
461 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= blocksperfile
* 1024
462 : blocksthisvol
>= blocksperfile
) {
467 else if (asize
> tsize
) {
480 struct slave_results returned
;
482 for (f
= 0; f
< SLAVES
; f
++) {
484 * Drain the results, but unlike EOT we DO (or should) care
485 * what the return values were, since if we detect EOT after
486 * we think we've written the last blocks to the tape anyway,
487 * we have to replay those blocks with rollforward.
489 * fixme: punt for now.
491 if (slaves
[f
].sent
) {
492 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
493 != sizeof returned
) {
494 perror(" DUMP: error reading command pipe in master");
497 got
= returned
.unclen
;
498 bytes_written
+= returned
.clen
;
499 if (returned
.unclen
== returned
.clen
)
507 msg("EOT detected in last 2 tape records!\n");
508 msg("Use a longer tape, decrease the size estimate\n");
509 quit("or use no size estimate at all.\n");
512 (void) close(slaves
[f
].fd
);
514 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
519 msg("Closing %s\n", tape
);
524 while (rmtopen(tape
, 0) < 0)
531 (void) close(tapefd
);
533 while ((f
= OPEN(tape
, 0)) < 0)
539 if (eot_script
&& spcl
.c_type
!= TS_END
) {
540 msg("Launching %s\n", eot_script
);
541 eot_code
= system_command(eot_script
, tape
, tapeno
);
543 if (eot_code
!= 0 && eot_code
!= 1) {
544 msg("Dump aborted by the end of tape script\n");
556 if (nexttape
|| Mflag
|| (eot_code
== 0) )
559 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
560 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
562 while (!query("Is the new volume mounted and ready to go?"))
563 if (query("Do you want to abort?")) {
572 register struct req
*p
, *q
= NULL
, *prev
;
573 register struct slave
*tslp
;
574 int i
, size
, savedtapea
, got
;
575 union u_spcl
*ntb
, *otb
;
576 struct slave_results returned
;
581 tslp
= &slaves
[SLAVES
];
582 ntb
= (union u_spcl
*)tslp
->tblock
[1];
585 * Each of the N slaves should have requests that need to
586 * be replayed on the next tape. Use the extra slave buffers
587 * (slaves[SLAVES]) to construct request lists to be sent to
588 * each slave in turn.
590 for (i
= 0; i
< SLAVES
; i
++) {
592 otb
= (union u_spcl
*)slp
->tblock
;
595 * For each request in the current slave, copy it to tslp.
599 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
602 *ntb
++ = *otb
++; /* copy the datablock also */
607 quit("rollforward: protocol botch");
619 nextblock
= tslp
->tblock
;
620 savedtapea
= spcl
.c_tapea
;
621 spcl
.c_tapea
= slp
->tapea
;
623 spcl
.c_tapea
= savedtapea
;
624 lastspclrec
= savedtapea
- 1;
626 size
= (char *)ntb
- (char *)q
;
627 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
628 perror(" DUMP: error writing command pipe");
633 lastfirstrec
= slp
->firstrec
;
635 if (++slp
>= &slaves
[SLAVES
])
640 if (prev
->dblk
!= 0) {
642 * If the last one was a disk block, make the
643 * first of this one be the last bit of that disk
646 q
->dblk
= prev
->dblk
+
647 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
648 ntb
= (union u_spcl
*)tslp
->tblock
;
651 * It wasn't a disk block. Copy the data to its
652 * new location in the buffer.
655 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
656 ntb
= (union u_spcl
*)tslp
->tblock
[1];
660 nextblock
= slp
->tblock
;
663 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
664 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
671 * Clear the first slaves' response. One hopes that it
672 * worked ok, otherwise the tape is much too short!
675 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
676 != sizeof returned
) {
677 perror(" DUMP: error reading command pipe in master");
680 got
= returned
.unclen
;
681 bytes_written
+= returned
.clen
;
682 if (returned
.clen
== returned
.unclen
)
690 quit("EOT detected at start of the tape!\n");
696 if (spcl
.c_type
!= TS_END
) {
697 for (i
= 0; i
< spcl
.c_count
; i
++)
698 if (spcl
.c_addr
[i
] != 0)
702 slp
->firstrec
= lastfirstrec
+ ntrec
;
703 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
705 asize
+= tenths
+ returned
.clen
/ density
;
706 blockswritten
+= ntrec
;
707 blocksthisvol
+= ntrec
;
712 * We implement taking and restoring checkpoints on the tape level.
713 * When each tape is opened, a new process is created by forking; this
714 * saves all of the necessary context in the parent. The child
715 * continues the dump; the parent waits around, saving the context.
716 * If the child returns X_REWRITE, then it had problems writing that tape;
717 * this causes the parent to fork again, duplicating the context, and
718 * everything continues as if nothing had happened.
721 startnewtape(int top
)
732 sigaddset(&sigs
, SIGINT
);
733 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
734 #else /* __linux__ */
736 void (*interrupt_save
)();
738 sig_t interrupt_save
;
740 interrupt_save
= signal(SIGINT
, SIG_IGN
);
741 #endif /* __linux__ */
743 parentpid
= getpid();
744 tapea_volume
= spcl
.c_tapea
;
745 tapea_bytes
= bytes_written
;
746 tstart_volume
= time(NULL
);
750 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
752 (void)signal(SIGINT
, interrupt_save
);
755 * All signals are inherited...
759 msg("Context save fork fails in parent %d\n", parentpid
);
765 * save the context by waiting
766 * until the child doing all of the work returns.
767 * don't catch the interrupt
770 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
772 signal(SIGINT
, SIG_IGN
);
775 msg("Tape: %d; parent process: %d child process %d\n",
776 tapeno
+1, parentpid
, childpid
);
778 while ((waitpid
= wait(&status
)) != childpid
)
779 if (waitpid
!= rshpid
)
780 msg("Parent %d waiting for child %d has another child %d return\n",
781 parentpid
, childpid
, waitpid
);
783 msg("Child %d returns LOB status %o\n",
784 childpid
, status
&0xFF);
786 status
= (status
>> 8) & 0xFF;
790 msg("Child %d finishes X_FINOK\n", childpid
);
793 msg("Child %d finishes X_ABORT\n", childpid
);
796 msg("Child %d finishes X_REWRITE\n", childpid
);
799 msg("Child %d finishes unknown %d\n",
810 goto restore_check_point
;
812 msg("Bad return code from dump: %d\n", status
);
816 } else { /* we are the child; just continue */
818 sleep(4); /* allow time for parent's message to get out */
819 msg("Child on Tape %d has parent %d, my pid = %d\n",
820 tapeno
+1, parentpid
, getpid());
823 * If we have a name like "/dev/rmt0,/dev/rmt1",
824 * use the name before the comma first, and save
825 * the remaining names for subsequent volumes.
827 tapeno
++; /* current tape sequence */
829 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
830 tape
[MAXPATHLEN
- 1] = '\0';
831 msg("Dumping volume %d on %s\n", tapeno
, tape
);
833 else if (nexttape
|| strchr(tapeprefix
, ',')) {
834 if (nexttape
&& *nexttape
)
835 tapeprefix
= nexttape
;
836 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
841 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
842 tape
[MAXPATHLEN
- 1] = '\0';
843 msg("Dumping volume %d on %s\n", tapeno
, tape
);
846 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
848 OPEN(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
850 while ((tapefd
= (pipeout
? fileno(stdout
) :
851 OPEN(tape
, O_RDWR
|O_CREAT
, 0666))) < 0)
854 msg("Cannot open output \"%s\".\n", tape
);
855 if (!query("Do you want to retry the open?"))
862 struct mtget mt_stat
;
863 magtapeout
= ioctl(tapefd
, MTIOCGET
, (char *)&mt_stat
) == 0;
865 msg("Output is to %s\n",
866 magtapeout ? "tape" : "file/pipe");
870 enslave(); /* Share open tape file descriptor with slaves */
875 newtape
++; /* new tape signal */
876 spcl
.c_count
= slp
->count
;
878 * measure firstrec in TP_BSIZE units since restore doesn't
879 * know the correct ntrec value...
881 spcl
.c_firstrec
= slp
->firstrec
;
883 spcl
.c_type
= TS_TAPE
;
884 spcl
.c_flags
|= DR_NEWHEADER
;
885 spcl
.c_ntrec
= ntrec
;
887 spcl
.c_flags
|= DR_COMPRESSED
;
888 writeheader((dump_ino_t
)slp
->inode
);
889 spcl
.c_flags
&=~ DR_NEWHEADER
;
890 msg("Volume %d started with block %ld at: %s", tapeno
,
891 spcl
.c_tapea
, ctime(&tstart_volume
));
893 msg("Volume %d begins with blocks from inode %d\n",
902 if (master
!= 0 && master
!= getpid())
903 /* Signals master to call dumpabort */
904 (void) kill(master
, SIGTERM
);
907 msg("The ENTIRE dump is aborted.\n");
920 msg("pid = %d exits with status %d\n", getpid(), status
);
926 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
932 siglongjmp(jmpbuf
, 1);
940 #ifdef LINUX_FORK_BUG
948 { struct sigaction sa
;
949 memset(&sa
, 0, sizeof sa
);
950 sigemptyset(&sa
.sa_mask
);
951 sa
.sa_handler
= dumpabort
;
952 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
953 sa
.sa_handler
= sigpipe
;
954 sigaction(SIGPIPE
, &sa
, NULL
);
955 sa
.sa_handler
= proceed
;
956 sa
.sa_flags
= SA_RESTART
;
957 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
960 for (i
= 0; i
< SLAVES
; i
++) {
961 if (i
== slp
- &slaves
[0]) {
967 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
968 (slaves
[i
].pid
= fork()) < 0)
969 quit("too many slaves, %d (recompile smaller): %s\n",
972 slaves
[i
].fd
= cmd
[1];
974 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
976 for (j
= 0; j
<= i
; j
++)
977 (void) close(slaves
[j
].fd
);
979 sigaddset(&sigs
, SIGINT
); /* Master handles this */
981 sigaddset(&sigs
, SIGINFO
);
983 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
985 #ifdef LINUX_FORK_BUG
986 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
988 quit("master/slave protocol botched 3\n");
990 doslave(cmd
[0], i
, (slaves
[i
].pid
== slp
->pid
));
997 #ifdef LINUX_FORK_BUG
999 * Wait for all slaves to _actually_ start to circumvent a bug in
1000 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1001 * returned from fork() causes a SEGV in the child process
1003 for (i
= 0; i
< SLAVES
; i
++)
1004 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1005 quit("master/slave protocol botched 4\n");
1008 for (i
= 0; i
< SLAVES
; i
++)
1009 (void) atomic_write( slaves
[i
].fd
,
1010 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1011 sizeof slaves
[0].pid
);
1021 for (i
= 0; i
< SLAVES
; i
++)
1022 if (slaves
[i
].pid
> 0) {
1023 (void) kill(slaves
[i
].pid
, SIGKILL
);
1029 * Synchronization - each process waits for a SIGUSR2 from the
1030 * previous process before writing to the tape, and sends SIGUSR2
1031 * to the next process when the tape write completes. On tape errors
1032 * a SIGUSR1 is sent to the master which then terminates all of the
1036 doslave(int cmd
, int slave_number
, int first
)
1039 int nextslave
, size
, eot_count
, bufsize
;
1040 volatile int wrote
= 0;
1042 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1043 struct tapebuf
*comp_buf
= NULL
;
1044 int compresult
, do_compress
= !first
;
1045 unsigned long worklen
;
1047 unsigned int worklen2
;
1049 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1050 struct slave_results returns
;
1056 union u_spcl
*uspclptr
;
1057 struct s_spcl
*spclptr
;
1058 #endif /* USE_QFA */
1062 sigaddset(&set
, SIGUSR2
);
1063 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1067 * Need our own seek pointer.
1069 (void) close(diskfd
);
1070 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1071 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1074 (void)ioctl(diskfd
, BLKFLSBUF
);
1077 retval
= dump_fs_open(disk
, &fs
);
1079 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1080 #endif /* __linux__ */
1083 * Need the pid of the next slave in the loop...
1085 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1086 != sizeof nextslave
) {
1087 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1090 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1091 /* if we're doing a compressed dump, allocate the compress buffer */
1093 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1094 if (comp_buf
== NULL
)
1095 quit("couldn't allocate a compress buffer.\n");
1097 comp_buf
->flags
= COMPRESS_BZLIB
;
1099 comp_buf
->flags
= COMPRESS_ZLIB
;
1101 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1104 * Get list of blocks to dump, read the blocks into tape buffer
1106 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1107 register struct req
*p
= slp
->req
;
1109 for (trecno
= 0; trecno
< ntrec
;
1110 trecno
+= p
->count
, p
+= p
->count
) {
1111 if (p
->dblk
) { /* read a disk block */
1112 bread(p
->dblk
, slp
->tblock
[trecno
],
1113 p
->count
* TP_BSIZE
);
1114 } else { /* read record from pipe */
1115 if (p
->count
!= 1 || atomic_read( cmd
,
1116 (char *)slp
->tblock
[trecno
],
1117 TP_BSIZE
) != TP_BSIZE
)
1118 quit("master/slave protocol botched.\n");
1122 /* Try to write the data... */
1126 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1127 bufsize
= writesize
; /* length to write */
1128 returns
.clen
= returns
.unclen
= bufsize
;
1130 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1132 * When writing a compressed dump, each block except
1133 * the first one on each tape is written
1134 * from struct tapebuf with an 4 byte prefix
1135 * followed by the data. This can be less than
1136 * writesize. Restore, on a short read, can compare the
1137 * length read to the compressed length in the header
1138 * to verify that the read was good. Blocks which don't
1139 * compress well are written uncompressed.
1140 * The first block written by each slave is not compressed
1141 * and does not have a prefix.
1144 if (compressed
&& do_compress
) {
1145 comp_buf
->length
= bufsize
;
1146 worklen
= TP_BSIZE
+ writesize
;
1150 compresult
= compress2(comp_buf
->buf
,
1152 (char *)slp
->tblock
[0],
1155 if (compresult
== Z_OK
)
1160 #endif /* HAVE_ZLIB */
1164 compresult
= BZ2_bzBuffToBuffCompress(
1167 (char *)slp
->tblock
[0],
1172 if (compresult
== BZ_OK
)
1178 #endif /* HAVE_BZLIB */
1179 if (compresult
&& worklen
<= (writesize
- 16)) {
1180 /* write the compressed buffer */
1181 comp_buf
->length
= worklen
;
1182 comp_buf
->compressed
= 1;
1183 buffer
= (char *) comp_buf
;
1184 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1187 /* write the data uncompressed */
1188 comp_buf
->length
= writesize
;
1189 comp_buf
->compressed
= 0;
1190 buffer
= (char *) comp_buf
;
1191 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1192 returns
.unclen
= returns
.clen
;
1193 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1196 /* compress the remaining blocks if we're compressing */
1197 do_compress
= compressed
;
1198 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1200 if (sigsetjmp(jmpbuf
, 1) == 0) {
1209 if (gTapeposfd
>= 0) {
1211 for (i
= 0; i
< ntrec
; ++i
) {
1212 uspclptr
= (union u_spcl
*)&slp
->tblock
[i
];
1213 spclptr
= &uspclptr
->s_spcl
;
1214 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1215 (spclptr
->c_type
== TS_INODE
) &&
1216 (spclptr
->c_date
== gThisDumpDate
)) {
1217 /* if an error occured previously don't
1220 if ((gtperr
= GetTapePos(&curtapepos
)) == 0)
1221 MkTapeString(spclptr
, curtapepos
);
1226 #endif /* USE_QFA */
1228 while (eot_count
< 10 && size
< bufsize
) {
1231 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1234 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1236 printf("slave %d wrote %d\n", slave_number
, wrote
);
1246 if (size
!= bufsize
)
1247 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1248 slave_number
, size
, bufsize
);
1252 * Handle ENOSPC as an EOT condition.
1254 if (wrote
< 0 && errno
== ENOSPC
) {
1260 returns
.clen
= returns
.unclen
= 0;
1263 * pass errno back to master for special handling
1266 returns
.unclen
= -errno
;
1269 * pass size of data and size of write back to master
1270 * (for EOT handling)
1272 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1275 * Signal the next slave to go.
1277 (void) kill(nextslave
, SIGUSR2
);
1280 quit("error reading command pipe: %s\n", strerror(errno
));
1284 * Since a read from a pipe may not return all we asked for,
1285 * or a write may not write all we ask if we get a signal,
1286 * loop until the count is satisfied (or error).
1289 atomic_read(int fd
, void *buf
, size_t count
)
1291 int got
, need
= count
;
1294 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1296 } while (got
== -1 && errno
== EINTR
);
1297 return (got
< 0 ? got
: count
- need
);
1301 * Since a read from a pipe may not return all we asked for,
1302 * or a write may not write all we ask if we get a signal,
1303 * loop until the count is satisfied (or error).
1306 atomic_write(int fd
, const void *buf
, size_t count
)
1308 int got
, need
= count
;
1311 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1313 } while (got
== -1 && errno
== EINTR
);
1314 return (got
< 0 ? got
: count
- need
);
1320 * read the current tape position
1323 GetTapePos(long *pos
)
1329 err
= (ioctl(tapefd
, MTIOCPOS
, pos
) < 0);
1332 *pos
= lseek(tapefd
, 0, SEEK_CUR
);
1337 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1345 MkTapeString(struct s_spcl
*spclptr
, long curtapepos
) {
1348 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1351 snprintf(gTps
, sizeof(gTps
), "%ld\t%d\t%ld\n",
1352 (unsigned long)spclptr
->c_inumber
,
1355 gTps
[sizeof(gTps
) - 1] = '\0';
1356 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1357 warn("error writing tapepos file.\n");
1360 #endif /* USE_QFA */