]>
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 1.33 2001/03/18 15:35:44 stelian Exp $";
49 #include <sys/types.h>
51 #include <linux/types.h>
53 #include <sys/param.h>
54 #include <sys/socket.h>
58 #include <linux/ext2_fs.h>
59 #include <bsdcompat.h>
62 #include <sys/vnode.h>
65 #include <ufs/inode.h>
67 #include <ufs/ufs/dinode.h>
68 #include <ufs/ffs/fs.h>
70 #endif /* __linux__ */
72 #include <protocols/dumprestore.h>
79 #include <compaterr.h>
89 #include <ext2fs/ext2fs.h>
94 #endif /* HAVE_ZLIB */
98 int writesize
; /* size of malloc()ed buffer for tape */
99 long lastspclrec
= -1; /* tape block number of last written header */
100 int trecno
= 0; /* next record to write in current block */
101 extern long blocksperfile
; /* number of blocks per output file */
102 long blocksthisvol
; /* number of blocks on current output file */
103 extern int ntrec
; /* blocking factor on tape */
104 extern int cartridge
;
109 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
111 static ssize_t atomic_read
__P((int, void *, size_t));
112 static ssize_t atomic_write
__P((int, const void *, size_t));
113 static void doslave
__P((int, int));
114 static void enslave
__P((void));
115 static void flushtape
__P((void));
116 static void killall
__P((void));
117 static void rollforward
__P((void));
118 static int system_command
__P((const char *, const char *, int));
121 * Concurrent dump mods (Caltech) - disk block reading and tape writing
122 * are exported to several slave processes. While one slave writes the
123 * tape, the others read disk blocks; they pass control of the tape in
124 * a ring via signals. The parent process traverses the filesystem and
125 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
126 * The following structure defines the instruction packets sent to slaves.
134 struct slave_results
{
135 ssize_t unclen
; /* uncompressed length */
136 ssize_t clen
; /* compressed length */
139 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
141 int tapea
; /* header number at start of this chunk */
142 int count
; /* count to next header (used for TS_TAPE */
144 int inode
; /* inode that we are currently dealing with */
145 int fd
; /* FD for this slave */
146 int pid
; /* PID for this slave */
147 int sent
; /* 1 == we've sent this slave requests */
148 int firstrec
; /* record number of this block */
149 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
150 struct req
*req
; /* buffer for requests */
154 char (*nextblock
)[TP_BSIZE
];
156 static time_t tstart_volume
; /* time of volume start */
157 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
159 int master
; /* pid of master, for sending error signals */
160 int tenths
; /* length of tape overhead per block written */
161 static int caught
; /* have we caught the signal to proceed? */
162 static int ready
; /* have we reached the lock point without having */
163 /* received the SIGUSR2 signal from the prev slave? */
164 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
165 /* SIGUSR2 arrives from the previous slave */
170 int pgoff
= getpagesize() - 1;
174 writesize
= ntrec
* TP_BSIZE
;
175 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
177 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
178 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
179 * repositioning after stopping, i.e, streaming mode, where the gap is
180 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
182 if (blocksperfile
== 0 && !unlimited
)
183 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
189 * Allocate tape buffer contiguous with the array of instruction
190 * packets, so flushtape() can write them together with one write().
191 * Align tape buffer on page boundary to speed up tape write().
193 for (i
= 0; i
<= SLAVES
; i
++) {
195 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
198 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
200 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
202 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
204 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
210 nextblock
= slp
->tblock
;
215 writerec(const void *dp
, int isspcl
)
218 slp
->req
[trecno
].dblk
= (daddr_t
)0;
219 slp
->req
[trecno
].count
= 1;
220 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
221 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
224 lastspclrec
= spcl
.c_tapea
;
232 dumpblock(daddr_t blkno
, int size
)
234 int avail
, tpblks
, dblkno
;
236 dblkno
= fsbtodb(sblock
, blkno
);
237 tpblks
= size
>> tp_bshift
;
238 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
239 slp
->req
[trecno
].dblk
= dblkno
;
240 slp
->req
[trecno
].count
= avail
;
242 spcl
.c_tapea
+= avail
;
245 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
257 msg("write error on %s: %s\n", tape
, strerror(errnum
));
258 quit("Cannot recover\n");
261 msg("write error %d blocks into volume %d: %s\n", 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
298 ttaken
= tnow
- tstart_volume
;
299 blocks
= spcl
.c_tapea
- tapea_volume
;
300 msg("Volume %d completed at: %s", tapeno
,
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,"
320 " compression ratio %1.3f\n",
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
)
365 int save_errno
= errno
;
371 (void) time((time_t *) &tnow
);
373 buf
= mktimeest(tnow
);
375 write(STDERR_FILENO
, buf
, strlen(buf
));
385 struct slave_results returned
;
387 int siz
= (char *)nextblock
- (char *)slp
->req
;
389 slp
->req
[trecno
].count
= 0; /* Sentinel */
391 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
392 quit("error writing command pipe: %s\n", strerror(errno
));
393 slp
->sent
= 1; /* we sent a request, read the response later */
395 lastfirstrec
= slp
->firstrec
;
397 if (++slp
>= &slaves
[SLAVES
])
400 /* Read results back from next slave */
402 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
403 != sizeof returned
) {
404 perror(" DUMP: error reading command pipe in master");
407 got
= returned
.unclen
;
408 bytes_written
+= returned
.clen
;
409 if (returned
.unclen
== returned
.clen
)
413 /* Check for errors */
417 /* Check for end of tape */
419 msg("End of tape detected\n");
422 * Drain the results, don't care what the values were.
423 * If we read them here then trewind won't...
425 for (i
= 0; i
< SLAVES
; i
++) {
426 if (slaves
[i
].sent
) {
427 if (atomic_read( slaves
[i
].fd
,
428 (char *)&returned
, sizeof returned
)
429 != sizeof returned
) {
430 perror(" DUMP: error reading command pipe in master");
444 if (spcl
.c_type
!= TS_END
) {
445 for (i
= 0; i
< spcl
.c_count
; i
++)
446 if (spcl
.c_addr
[i
] != 0)
449 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
450 slp
->tapea
= spcl
.c_tapea
;
451 slp
->firstrec
= lastfirstrec
+ ntrec
;
453 nextblock
= slp
->tblock
;
455 asize
+= tenths
+ returned
.clen
/ density
;
456 blockswritten
+= ntrec
;
457 blocksthisvol
+= ntrec
;
458 if (!pipeout
&& !unlimited
&& (blocksperfile
?
459 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
467 * Executes the command in a shell.
468 * Returns -1 if an error occured, the exit status of
469 * the command on success.
471 int system_command(const char *command
, const char *device
, int volnum
) {
473 char commandstr
[4096];
477 perror(" DUMP: unable to fork");
483 #if OLD_STYLE_FSCRIPT
484 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
486 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
488 commandstr
[sizeof(commandstr
) - 1] = '\0';
489 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
490 perror(" DUMP: unable to execute shell");
494 if (waitpid(pid
, &status
, 0) == -1) {
495 if (errno
!= EINTR
) {
496 perror(" DUMP: waitpid error");
500 if (WIFEXITED(status
))
501 return WEXITSTATUS(status
);
513 struct slave_results returned
;
515 for (f
= 0; f
< SLAVES
; f
++) {
517 * Drain the results, but unlike EOT we DO (or should) care
518 * what the return values were, since if we detect EOT after
519 * we think we've written the last blocks to the tape anyway,
520 * we have to replay those blocks with rollforward.
522 * fixme: punt for now.
524 if (slaves
[f
].sent
) {
525 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
526 != sizeof returned
) {
527 perror(" DUMP: error reading command pipe in master");
530 got
= returned
.unclen
;
531 bytes_written
+= returned
.clen
;
532 if (returned
.unclen
== returned
.clen
)
540 msg("EOT detected in last 2 tape records!\n");
541 msg("Use a longer tape, decrease the size estimate\n");
542 quit("or use no size estimate at all.\n");
545 (void) close(slaves
[f
].fd
);
547 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
552 msg("Closing %s\n", tape
);
557 while (rmtopen(tape
, 0) < 0)
564 (void) close(tapefd
);
565 while ((f
= open(tape
, 0)) < 0)
570 if (eot_script
&& spcl
.c_type
!= TS_END
) {
571 msg("Launching %s\n", eot_script
);
572 eot_code
= system_command(eot_script
, tape
, tapeno
);
574 if (eot_code
!= 0 && eot_code
!= 1) {
575 msg("Dump aborted by the end of tape script\n");
587 if (nexttape
|| Mflag
|| (eot_code
== 0) )
590 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
591 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
593 while (!query("Is the new volume mounted and ready to go?"))
594 if (query("Do you want to abort?")) {
603 register struct req
*p
, *q
, *prev
;
604 register struct slave
*tslp
;
605 int i
, size
, savedtapea
, got
;
606 union u_spcl
*ntb
, *otb
;
607 struct slave_results returned
;
612 tslp
= &slaves
[SLAVES
];
613 ntb
= (union u_spcl
*)tslp
->tblock
[1];
616 * Each of the N slaves should have requests that need to
617 * be replayed on the next tape. Use the extra slave buffers
618 * (slaves[SLAVES]) to construct request lists to be sent to
619 * each slave in turn.
621 for (i
= 0; i
< SLAVES
; i
++) {
623 otb
= (union u_spcl
*)slp
->tblock
;
626 * For each request in the current slave, copy it to tslp.
630 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
633 *ntb
++ = *otb
++; /* copy the datablock also */
638 quit("rollforward: protocol botch");
650 nextblock
= tslp
->tblock
;
651 savedtapea
= spcl
.c_tapea
;
652 spcl
.c_tapea
= slp
->tapea
;
654 spcl
.c_tapea
= savedtapea
;
655 lastspclrec
= savedtapea
- 1;
657 size
= (char *)ntb
- (char *)q
;
658 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
659 perror(" DUMP: error writing command pipe");
664 lastfirstrec
= slp
->firstrec
;
666 if (++slp
>= &slaves
[SLAVES
])
671 if (prev
->dblk
!= 0) {
673 * If the last one was a disk block, make the
674 * first of this one be the last bit of that disk
677 q
->dblk
= prev
->dblk
+
678 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
679 ntb
= (union u_spcl
*)tslp
->tblock
;
682 * It wasn't a disk block. Copy the data to its
683 * new location in the buffer.
686 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
687 ntb
= (union u_spcl
*)tslp
->tblock
[1];
691 nextblock
= slp
->tblock
;
694 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
695 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
702 * Clear the first slaves' response. One hopes that it
703 * worked ok, otherwise the tape is much too short!
706 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
707 != sizeof returned
) {
708 perror(" DUMP: error reading command pipe in master");
711 got
= returned
.unclen
;
712 bytes_written
+= returned
.clen
;
713 if (returned
.clen
== returned
.unclen
)
721 quit("EOT detected at start of the tape!\n");
727 if (spcl
.c_type
!= TS_END
) {
728 for (i
= 0; i
< spcl
.c_count
; i
++)
729 if (spcl
.c_addr
[i
] != 0)
733 slp
->firstrec
= lastfirstrec
+ ntrec
;
734 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
736 asize
+= tenths
+ returned
.clen
/ density
;
737 blockswritten
+= ntrec
;
738 blocksthisvol
+= ntrec
;
743 * We implement taking and restoring checkpoints on the tape level.
744 * When each tape is opened, a new process is created by forking; this
745 * saves all of the necessary context in the parent. The child
746 * continues the dump; the parent waits around, saving the context.
747 * If the child returns X_REWRITE, then it had problems writing that tape;
748 * this causes the parent to fork again, duplicating the context, and
749 * everything continues as if nothing had happened.
752 startnewtape(int top
)
763 sigaddset(&sigs
, SIGINT
);
764 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
765 #else /* __linux__ */
767 void (*interrupt_save
)();
769 sig_t interrupt_save
;
771 interrupt_save
= signal(SIGINT
, SIG_IGN
);
772 #endif /* __linux__ */
774 parentpid
= getpid();
775 tapea_volume
= spcl
.c_tapea
;
776 tapea_bytes
= bytes_written
;
778 (void)time4(&tstart_volume
);
780 (void)time((&tstart_volume
);
785 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
787 (void)signal(SIGINT
, interrupt_save
);
790 * All signals are inherited...
794 msg("Context save fork fails in parent %d\n", parentpid
);
800 * save the context by waiting
801 * until the child doing all of the work returns.
802 * don't catch the interrupt
805 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
807 signal(SIGINT
, SIG_IGN
);
810 msg("Tape: %d; parent process: %d child process %d\n",
811 tapeno
+1, parentpid
, childpid
);
813 while ((waitpid
= wait(&status
)) != childpid
)
814 if (waitpid
!= rshpid
)
815 msg("Parent %d waiting for child %d has another child %d return\n",
816 parentpid
, childpid
, waitpid
);
818 msg("Child %d returns LOB status %o\n",
819 childpid
, status
&0xFF);
821 status
= (status
>> 8) & 0xFF;
825 msg("Child %d finishes X_FINOK\n", childpid
);
828 msg("Child %d finishes X_ABORT\n", childpid
);
831 msg("Child %d finishes X_REWRITE\n", childpid
);
834 msg("Child %d finishes unknown %d\n",
845 goto restore_check_point
;
847 msg("Bad return code from dump: %d\n", status
);
851 } else { /* we are the child; just continue */
853 sleep(4); /* allow time for parent's message to get out */
854 msg("Child on Tape %d has parent %d, my pid = %d\n",
855 tapeno
+1, parentpid
, getpid());
858 * If we have a name like "/dev/rmt0,/dev/rmt1",
859 * use the name before the comma first, and save
860 * the remaining names for subsequent volumes.
862 tapeno
++; /* current tape sequence */
864 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
865 tape
[MAXPATHLEN
- 1] = '\0';
866 msg("Dumping volume %d on %s\n", tapeno
, tape
);
868 else if (nexttape
|| strchr(tapeprefix
, ',')) {
869 if (nexttape
&& *nexttape
)
870 tapeprefix
= nexttape
;
871 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
876 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
877 tape
[MAXPATHLEN
- 1] = '\0';
878 msg("Dumping volume %d on %s\n", tapeno
, tape
);
881 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
883 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
885 while ((tapefd
= (pipeout
? fileno(stdout
) :
886 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
889 msg("Cannot open output \"%s\".\n", tape
);
890 if (!query("Do you want to retry the open?"))
894 enslave(); /* Share open tape file descriptor with slaves */
899 newtape
++; /* new tape signal */
900 spcl
.c_count
= slp
->count
;
902 * measure firstrec in TP_BSIZE units since restore doesn't
903 * know the correct ntrec value...
905 spcl
.c_firstrec
= slp
->firstrec
;
907 spcl
.c_type
= TS_TAPE
;
908 spcl
.c_flags
|= DR_NEWHEADER
;
910 spcl
.c_flags
|= DR_COMPRESSED
;
911 writeheader((ino_t
)slp
->inode
);
912 spcl
.c_flags
&=~ DR_NEWHEADER
;
913 msg("Volume %d started at: %s", tapeno
,
915 ctime4(&tstart_volume
));
917 ctime(&tstart_volume
));
920 msg("Volume %d begins with blocks from inode %d\n",
929 if (master
!= 0 && master
!= getpid())
930 /* Signals master to call dumpabort */
931 (void) kill(master
, SIGTERM
);
934 msg("The ENTIRE dump is aborted.\n");
947 msg("pid = %d exits with status %d\n", getpid(), status
);
953 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
959 siglongjmp(jmpbuf
, 1);
967 #ifdef LINUX_FORK_BUG
975 { struct sigaction sa
;
976 memset(&sa
, 0, sizeof sa
);
977 sigemptyset(&sa
.sa_mask
);
978 sa
.sa_handler
= dumpabort
;
979 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
980 sa
.sa_handler
= sigpipe
;
981 sigaction(SIGPIPE
, &sa
, NULL
);
982 sa
.sa_handler
= proceed
;
983 sa
.sa_flags
= SA_RESTART
;
984 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
987 for (i
= 0; i
< SLAVES
; i
++) {
988 if (i
== slp
- &slaves
[0]) {
994 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
995 (slaves
[i
].pid
= fork()) < 0)
996 quit("too many slaves, %d (recompile smaller): %s\n",
999 slaves
[i
].fd
= cmd
[1];
1001 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
1003 for (j
= 0; j
<= i
; j
++)
1004 (void) close(slaves
[j
].fd
);
1006 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1007 #if defined(SIGINFO)
1008 sigaddset(&sigs
, SIGINFO
);
1010 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1012 #ifdef LINUX_FORK_BUG
1013 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1015 quit("master/slave protocol botched 3\n");
1022 #ifdef LINUX_FORK_BUG
1024 * Wait for all slaves to _actually_ start to circumvent a bug in
1025 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1026 * returned from fork() causes a SEGV in the child process
1028 for (i
= 0; i
< SLAVES
; i
++)
1029 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1030 quit("master/slave protocol botched 4\n");
1033 for (i
= 0; i
< SLAVES
; i
++)
1034 (void) atomic_write( slaves
[i
].fd
,
1035 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1036 sizeof slaves
[0].pid
);
1046 for (i
= 0; i
< SLAVES
; i
++)
1047 if (slaves
[i
].pid
> 0) {
1048 (void) kill(slaves
[i
].pid
, SIGKILL
);
1054 * Synchronization - each process waits for a SIGUSR2 from the
1055 * previous process before writing to the tape, and sends SIGUSR2
1056 * to the next process when the tape write completes. On tape errors
1057 * a SIGUSR1 is sent to the master which then terminates all of the
1061 doslave(int cmd
, int slave_number
)
1064 int nextslave
, size
, eot_count
, bufsize
;
1065 volatile int wrote
= 0;
1068 struct tapebuf
*comp_buf
= NULL
;
1069 int compresult
, complevel
= 6, do_compress
= 0;
1070 unsigned long worklen
;
1071 #endif /* HAVE_ZLIB */
1072 struct slave_results returns
;
1078 * Need our own seek pointer.
1080 (void) close(diskfd
);
1081 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
1082 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1085 retval
= dump_fs_open(disk
, &fs
);
1087 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1088 #endif /* __linux__ */
1091 * Need the pid of the next slave in the loop...
1093 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1094 != sizeof nextslave
) {
1095 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1099 /* if we're doing a compressed dump, allocate the compress buffer */
1101 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1102 if (comp_buf
== NULL
)
1103 quit("couldn't allocate a compress buffer.\n");
1104 comp_buf
->flags
= 0;
1106 #endif /* HAVE_ZLIB */
1109 * Get list of blocks to dump, read the blocks into tape buffer
1111 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1112 register struct req
*p
= slp
->req
;
1114 for (trecno
= 0; trecno
< ntrec
;
1115 trecno
+= p
->count
, p
+= p
->count
) {
1116 if (p
->dblk
) { /* read a disk block */
1117 bread(p
->dblk
, slp
->tblock
[trecno
],
1118 p
->count
* TP_BSIZE
);
1119 } else { /* read record from pipe */
1120 if (p
->count
!= 1 || atomic_read( cmd
,
1121 (char *)slp
->tblock
[trecno
],
1122 TP_BSIZE
) != TP_BSIZE
)
1123 quit("master/slave protocol botched.\n");
1126 if (sigsetjmp(jmpbuf
, 1) == 0) {
1134 /* Try to write the data... */
1138 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1139 bufsize
= writesize
; /* length to write */
1140 returns
.clen
= returns
.unclen
= bufsize
;
1144 * The first NR_SLAVE blocks are not compressed.
1145 * When writing a compressed dump, each block is
1146 * written from struct tapebuf with an 4 byte prefix
1147 * followed by the data. This can be less than
1148 * writesize. Restore, on a short read, can compare the
1149 * length read to the compressed length in the header
1150 * to verify that the read was good. Blocks which don't
1151 * compress well are written uncompressed.
1155 comp_buf
->length
= bufsize
;
1156 worklen
= TP_BSIZE
+ writesize
;
1158 compresult
= compress2(comp_buf
->buf
, &worklen
,
1159 (char *)slp
->tblock
[0], writesize
, complevel
);
1160 if (compresult
== Z_OK
&& worklen
<= writesize
-32) {
1161 /* write the compressed buffer */
1162 comp_buf
->length
= worklen
;
1163 comp_buf
->compressed
= 1;
1164 buffer
= (char *) comp_buf
;
1165 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1168 /* write the data uncompressed */
1169 comp_buf
->length
= writesize
;
1170 comp_buf
->compressed
= 0;
1171 buffer
= (char *) comp_buf
;
1172 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1173 returns
.unclen
= returns
.clen
;
1174 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1177 /* compress the remaining blocks */
1178 do_compress
= compressed
;
1179 #endif /* HAVE_ZLIB */
1181 while (eot_count
< 10 && size
< bufsize
) {
1184 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1187 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1189 printf("slave %d wrote %d\n", slave_number
, wrote
);
1199 if (size
!= bufsize
)
1200 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1201 slave_number
, size
, bufsize
);
1205 * Handle ENOSPC as an EOT condition.
1207 if (wrote
< 0 && errno
== ENOSPC
) {
1213 returns
.clen
= returns
.unclen
= 0;
1216 * pass errno back to master for special handling
1219 returns
.unclen
= -errno
;
1222 * pass size of data and size of write back to master
1223 * (for EOT handling)
1225 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1228 * Signal the next slave to go.
1230 (void) kill(nextslave
, SIGUSR2
);
1233 quit("error reading command pipe: %s\n", strerror(errno
));
1237 * Since a read from a pipe may not return all we asked for,
1238 * or a write may not write all we ask if we get a signal,
1239 * loop until the count is satisfied (or error).
1242 atomic_read(int fd
, void *buf
, size_t count
)
1244 int got
, need
= count
;
1247 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1249 } while (got
== -1 && errno
== EINTR
);
1250 return (got
< 0 ? got
: count
- need
);
1254 * Since a read from a pipe may not return all we asked for,
1255 * or a write may not write all we ask if we get a signal,
1256 * loop until the count is satisfied (or error).
1259 atomic_write(int fd
, const void *buf
, size_t count
)
1261 int got
, need
= count
;
1264 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1266 } while (got
== -1 && errno
== EINTR
);
1267 return (got
< 0 ? got
: count
- need
);