]>
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 1.49 2001/07/18 09:50:48 stelian Exp $";
48 #include <compatlfs.h>
54 #include <compaterr.h>
64 #include <sys/types.h>
68 #include <sys/param.h>
69 #include <sys/socket.h>
74 #ifdef HAVE_EXT2FS_EXT2_FS_H
75 #include <ext2fs/ext2_fs.h>
77 #include <linux/ext2_fs.h>
79 #include <ext2fs/ext2fs.h>
80 #include <bsdcompat.h>
82 #include <sys/vnode.h>
85 #include <ufs/inode.h>
87 #include <ufs/ufs/dinode.h>
88 #include <ufs/ffs/fs.h>
89 #endif /* __linux__ */
91 #include <protocols/dumprestore.h>
95 #endif /* HAVE_ZLIB */
99 int writesize
; /* size of malloc()ed buffer for tape */
100 long lastspclrec
= -1; /* tape block number of last written header */
101 int trecno
= 0; /* next record to write in current block */
102 extern long blocksperfile
; /* number of blocks per output file */
103 long blocksthisvol
; /* number of blocks on current output file */
104 extern int ntrec
; /* blocking factor on tape */
105 extern int cartridge
;
110 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
112 static ssize_t atomic_read
__P((int, void *, size_t));
113 static ssize_t atomic_write
__P((int, const void *, size_t));
114 static void doslave
__P((int, int));
115 static void enslave
__P((void));
116 static void flushtape
__P((void));
117 static void killall
__P((void));
118 static void rollforward
__P((void));
119 static int system_command
__P((const char *, const char *, int));
122 * Concurrent dump mods (Caltech) - disk block reading and tape writing
123 * are exported to several slave processes. While one slave writes the
124 * tape, the others read disk blocks; they pass control of the tape in
125 * a ring via signals. The parent process traverses the filesystem and
126 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
127 * The following structure defines the instruction packets sent to slaves.
135 struct slave_results
{
136 ssize_t unclen
; /* uncompressed length */
137 ssize_t clen
; /* compressed length */
140 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
142 int tapea
; /* header number at start of this chunk */
143 int count
; /* count to next header (used for TS_TAPE */
145 int inode
; /* inode that we are currently dealing with */
146 int fd
; /* FD for this slave */
147 int pid
; /* PID for this slave */
148 int sent
; /* 1 == we've sent this slave requests */
149 int firstrec
; /* record number of this block */
150 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
151 struct req
*req
; /* buffer for requests */
155 char (*nextblock
)[TP_BSIZE
];
157 static time_t tstart_volume
; /* time of volume start */
158 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
160 int master
; /* pid of master, for sending error signals */
161 int tenths
; /* length of tape overhead per block written */
162 static int caught
; /* have we caught the signal to proceed? */
163 static int ready
; /* have we reached the lock point without having */
164 /* received the SIGUSR2 signal from the prev slave? */
165 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
166 /* SIGUSR2 arrives from the previous slave */
168 static int gtperr
= 0;
174 int pgoff
= getpagesize() - 1;
178 writesize
= ntrec
* TP_BSIZE
;
179 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
181 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
182 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
183 * repositioning after stopping, i.e, streaming mode, where the gap is
184 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
186 if (blocksperfile
== 0 && !unlimited
)
187 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
193 * Allocate tape buffer contiguous with the array of instruction
194 * packets, so flushtape() can write them together with one write().
195 * Align tape buffer on page boundary to speed up tape write().
197 for (i
= 0; i
<= SLAVES
; i
++) {
199 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
202 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
204 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
206 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
208 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
214 nextblock
= slp
->tblock
;
219 writerec(const void *dp
, int isspcl
)
222 slp
->req
[trecno
].dblk
= (daddr_t
)0;
223 slp
->req
[trecno
].count
= 1;
224 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
225 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
228 lastspclrec
= spcl
.c_tapea
;
236 dumpblock(daddr_t blkno
, int size
)
238 int avail
, tpblks
, dblkno
;
240 dblkno
= fsbtodb(sblock
, blkno
);
241 tpblks
= size
>> tp_bshift
;
242 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
243 slp
->req
[trecno
].dblk
= dblkno
;
244 slp
->req
[trecno
].count
= avail
;
246 spcl
.c_tapea
+= avail
;
249 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
261 msg("write error on %s: %s\n", tape
, strerror(errnum
));
262 quit("Cannot recover\n");
265 msg("write error %d blocks into volume %d: %s\n",
266 blocksthisvol
, tapeno
, strerror(errnum
));
267 broadcast("DUMP WRITE ERROR!\n");
268 if (query("Do you want to rewrite this volume?")) {
269 msg("Closing this volume. Prepare to restart with new media;\n");
270 msg("this dump volume will be rewritten.\n");
276 if (query("Do you want to start the next tape?"))
285 quit("Broken pipe\n");
290 * Update xferrate stats
299 ttaken
= tnow
- tstart_volume
;
300 blocks
= spcl
.c_tapea
- tapea_volume
;
301 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
303 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
304 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
306 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
307 long txfrate
= volkb
/ ttaken
;
308 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
309 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
310 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
314 double rate
= .0005 + (double) blocks
/ (double) volkb
;
315 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
317 tapeno
, blocks
, volkb
, rate
);
324 mktimeest(time_t tnow
)
326 static char msgbuf
[128];
331 if (blockswritten
< 500)
333 if (blockswritten
> tapesize
)
334 tapesize
= blockswritten
;
335 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
336 / blockswritten
* tapesize
;
337 if (tnow
> tstart_volume
)
338 (void)snprintf(msgbuf
, sizeof(msgbuf
),
339 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
340 (blockswritten
* 100.0) / tapesize
,
341 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
342 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
344 (void)snprintf(msgbuf
, sizeof(msgbuf
),
345 "%3.2f%% done, finished in %d:%02d\n",
346 (blockswritten
* 100.0) / tapesize
,
347 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
355 * information message upon receipt of SIGINFO
358 statussig(int notused
)
360 int save_errno
= errno
;
363 buf
= mktimeest(time(NULL
));
365 write(STDERR_FILENO
, buf
, strlen(buf
));
375 struct slave_results returned
;
377 int siz
= (char *)nextblock
- (char *)slp
->req
;
379 slp
->req
[trecno
].count
= 0; /* Sentinel */
381 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
382 quit("error writing command pipe: %s\n", strerror(errno
));
383 slp
->sent
= 1; /* we sent a request, read the response later */
385 lastfirstrec
= slp
->firstrec
;
387 if (++slp
>= &slaves
[SLAVES
])
390 /* Read results back from next slave */
392 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
393 != sizeof returned
) {
394 perror(" DUMP: error reading command pipe in master");
397 got
= returned
.unclen
;
398 bytes_written
+= returned
.clen
;
399 if (returned
.unclen
== returned
.clen
)
403 /* Check for errors or end of tape */
405 /* Check for errors */
409 msg("End of tape detected\n");
412 * Drain the results, don't care what the values were.
413 * If we read them here then trewind won't...
415 for (i
= 0; i
< SLAVES
; i
++) {
416 if (slaves
[i
].sent
) {
417 if (atomic_read( slaves
[i
].fd
,
418 (char *)&returned
, sizeof returned
)
419 != sizeof returned
) {
420 perror(" DUMP: error reading command pipe in master");
434 if (spcl
.c_type
!= TS_END
) {
435 for (i
= 0; i
< spcl
.c_count
; i
++)
436 if (spcl
.c_addr
[i
] != 0)
439 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
440 slp
->tapea
= spcl
.c_tapea
;
441 slp
->firstrec
= lastfirstrec
+ ntrec
;
443 nextblock
= slp
->tblock
;
445 asize
+= tenths
+ returned
.clen
/ density
;
446 csize
+= returned
.clen
;
447 blockswritten
+= ntrec
;
448 blocksthisvol
+= ntrec
;
449 if (!pipeout
&& !unlimited
) {
451 if ( compressed
? csize
>= blocksperfile
* 1024
452 : blocksthisvol
>= blocksperfile
) {
457 else if (asize
> tsize
) {
466 * Executes the command in a shell.
467 * Returns -1 if an error occured, the exit status of
468 * the command on success.
470 int system_command(const char *command
, const char *device
, int volnum
) {
472 char commandstr
[4096];
476 perror(" DUMP: unable to fork");
482 #if OLD_STYLE_FSCRIPT
483 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
485 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
487 commandstr
[sizeof(commandstr
) - 1] = '\0';
488 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
489 perror(" DUMP: unable to execute shell");
493 if (waitpid(pid
, &status
, 0) == -1) {
494 if (errno
!= EINTR
) {
495 perror(" DUMP: waitpid error");
499 if (WIFEXITED(status
))
500 return WEXITSTATUS(status
);
512 struct slave_results returned
;
514 for (f
= 0; f
< SLAVES
; f
++) {
516 * Drain the results, but unlike EOT we DO (or should) care
517 * what the return values were, since if we detect EOT after
518 * we think we've written the last blocks to the tape anyway,
519 * we have to replay those blocks with rollforward.
521 * fixme: punt for now.
523 if (slaves
[f
].sent
) {
524 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
525 != sizeof returned
) {
526 perror(" DUMP: error reading command pipe in master");
529 got
= returned
.unclen
;
530 bytes_written
+= returned
.clen
;
531 if (returned
.unclen
== returned
.clen
)
539 msg("EOT detected in last 2 tape records!\n");
540 msg("Use a longer tape, decrease the size estimate\n");
541 quit("or use no size estimate at all.\n");
544 (void) close(slaves
[f
].fd
);
546 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
551 msg("Closing %s\n", tape
);
556 while (rmtopen(tape
, 0) < 0)
563 (void) close(tapefd
);
565 while ((f
= OPEN(tape
, 0)) < 0)
571 if (eot_script
&& spcl
.c_type
!= TS_END
) {
572 msg("Launching %s\n", eot_script
);
573 eot_code
= system_command(eot_script
, tape
, tapeno
);
575 if (eot_code
!= 0 && eot_code
!= 1) {
576 msg("Dump aborted by the end of tape script\n");
588 if (nexttape
|| Mflag
|| (eot_code
== 0) )
591 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
592 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
594 while (!query("Is the new volume mounted and ready to go?"))
595 if (query("Do you want to abort?")) {
604 register struct req
*p
, *q
, *prev
;
605 register struct slave
*tslp
;
606 int i
, size
, savedtapea
, got
;
607 union u_spcl
*ntb
, *otb
;
608 struct slave_results returned
;
613 tslp
= &slaves
[SLAVES
];
614 ntb
= (union u_spcl
*)tslp
->tblock
[1];
617 * Each of the N slaves should have requests that need to
618 * be replayed on the next tape. Use the extra slave buffers
619 * (slaves[SLAVES]) to construct request lists to be sent to
620 * each slave in turn.
622 for (i
= 0; i
< SLAVES
; i
++) {
624 otb
= (union u_spcl
*)slp
->tblock
;
627 * For each request in the current slave, copy it to tslp.
631 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
634 *ntb
++ = *otb
++; /* copy the datablock also */
639 quit("rollforward: protocol botch");
651 nextblock
= tslp
->tblock
;
652 savedtapea
= spcl
.c_tapea
;
653 spcl
.c_tapea
= slp
->tapea
;
655 spcl
.c_tapea
= savedtapea
;
656 lastspclrec
= savedtapea
- 1;
658 size
= (char *)ntb
- (char *)q
;
659 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
660 perror(" DUMP: error writing command pipe");
665 lastfirstrec
= slp
->firstrec
;
667 if (++slp
>= &slaves
[SLAVES
])
672 if (prev
->dblk
!= 0) {
674 * If the last one was a disk block, make the
675 * first of this one be the last bit of that disk
678 q
->dblk
= prev
->dblk
+
679 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
680 ntb
= (union u_spcl
*)tslp
->tblock
;
683 * It wasn't a disk block. Copy the data to its
684 * new location in the buffer.
687 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
688 ntb
= (union u_spcl
*)tslp
->tblock
[1];
692 nextblock
= slp
->tblock
;
695 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
696 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
703 * Clear the first slaves' response. One hopes that it
704 * worked ok, otherwise the tape is much too short!
707 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
708 != sizeof returned
) {
709 perror(" DUMP: error reading command pipe in master");
712 got
= returned
.unclen
;
713 bytes_written
+= returned
.clen
;
714 if (returned
.clen
== returned
.unclen
)
722 quit("EOT detected at start of the tape!\n");
728 if (spcl
.c_type
!= TS_END
) {
729 for (i
= 0; i
< spcl
.c_count
; i
++)
730 if (spcl
.c_addr
[i
] != 0)
734 slp
->firstrec
= lastfirstrec
+ ntrec
;
735 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
737 asize
+= tenths
+ returned
.clen
/ density
;
738 csize
+= returned
.clen
;
739 blockswritten
+= ntrec
;
740 blocksthisvol
+= ntrec
;
745 * We implement taking and restoring checkpoints on the tape level.
746 * When each tape is opened, a new process is created by forking; this
747 * saves all of the necessary context in the parent. The child
748 * continues the dump; the parent waits around, saving the context.
749 * If the child returns X_REWRITE, then it had problems writing that tape;
750 * this causes the parent to fork again, duplicating the context, and
751 * everything continues as if nothing had happened.
754 startnewtape(int top
)
765 sigaddset(&sigs
, SIGINT
);
766 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
767 #else /* __linux__ */
769 void (*interrupt_save
)();
771 sig_t interrupt_save
;
773 interrupt_save
= signal(SIGINT
, SIG_IGN
);
774 #endif /* __linux__ */
776 parentpid
= getpid();
777 tapea_volume
= spcl
.c_tapea
;
778 tapea_bytes
= bytes_written
;
779 tstart_volume
= time(NULL
);
783 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
785 (void)signal(SIGINT
, interrupt_save
);
788 * All signals are inherited...
792 msg("Context save fork fails in parent %d\n", parentpid
);
798 * save the context by waiting
799 * until the child doing all of the work returns.
800 * don't catch the interrupt
803 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
805 signal(SIGINT
, SIG_IGN
);
808 msg("Tape: %d; parent process: %d child process %d\n",
809 tapeno
+1, parentpid
, childpid
);
811 while ((waitpid
= wait(&status
)) != childpid
)
812 if (waitpid
!= rshpid
)
813 msg("Parent %d waiting for child %d has another child %d return\n",
814 parentpid
, childpid
, waitpid
);
816 msg("Child %d returns LOB status %o\n",
817 childpid
, status
&0xFF);
819 status
= (status
>> 8) & 0xFF;
823 msg("Child %d finishes X_FINOK\n", childpid
);
826 msg("Child %d finishes X_ABORT\n", childpid
);
829 msg("Child %d finishes X_REWRITE\n", childpid
);
832 msg("Child %d finishes unknown %d\n",
843 goto restore_check_point
;
845 msg("Bad return code from dump: %d\n", status
);
849 } else { /* we are the child; just continue */
851 sleep(4); /* allow time for parent's message to get out */
852 msg("Child on Tape %d has parent %d, my pid = %d\n",
853 tapeno
+1, parentpid
, getpid());
856 * If we have a name like "/dev/rmt0,/dev/rmt1",
857 * use the name before the comma first, and save
858 * the remaining names for subsequent volumes.
860 tapeno
++; /* current tape sequence */
862 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
863 tape
[MAXPATHLEN
- 1] = '\0';
864 msg("Dumping volume %d on %s\n", tapeno
, tape
);
866 else if (nexttape
|| strchr(tapeprefix
, ',')) {
867 if (nexttape
&& *nexttape
)
868 tapeprefix
= nexttape
;
869 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
874 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
875 tape
[MAXPATHLEN
- 1] = '\0';
876 msg("Dumping volume %d on %s\n", tapeno
, tape
);
879 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
881 OPEN(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
883 while ((tapefd
= (pipeout
? fileno(stdout
) :
884 OPEN(tape
, O_RDWR
|O_CREAT
, 0666))) < 0)
887 msg("Cannot open output \"%s\".\n", tape
);
888 if (!query("Do you want to retry the open?"))
892 enslave(); /* Share open tape file descriptor with slaves */
898 newtape
++; /* new tape signal */
899 spcl
.c_count
= slp
->count
;
901 * measure firstrec in TP_BSIZE units since restore doesn't
902 * know the correct ntrec value...
904 spcl
.c_firstrec
= slp
->firstrec
;
906 spcl
.c_type
= TS_TAPE
;
907 spcl
.c_flags
|= DR_NEWHEADER
;
908 spcl
.c_ntrec
= ntrec
;
910 spcl
.c_flags
|= DR_COMPRESSED
;
911 writeheader((dump_ino_t
)slp
->inode
);
912 spcl
.c_flags
&=~ DR_NEWHEADER
;
913 msg("Volume %d started with block %ld at: %s", tapeno
,
914 spcl
.c_tapea
, ctime(&tstart_volume
));
916 msg("Volume %d begins with blocks from inode %d\n",
925 if (master
!= 0 && master
!= getpid())
926 /* Signals master to call dumpabort */
927 (void) kill(master
, SIGTERM
);
930 msg("The ENTIRE dump is aborted.\n");
943 msg("pid = %d exits with status %d\n", getpid(), status
);
949 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
955 siglongjmp(jmpbuf
, 1);
963 #ifdef LINUX_FORK_BUG
971 { struct sigaction sa
;
972 memset(&sa
, 0, sizeof sa
);
973 sigemptyset(&sa
.sa_mask
);
974 sa
.sa_handler
= dumpabort
;
975 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
976 sa
.sa_handler
= sigpipe
;
977 sigaction(SIGPIPE
, &sa
, NULL
);
978 sa
.sa_handler
= proceed
;
979 sa
.sa_flags
= SA_RESTART
;
980 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
983 for (i
= 0; i
< SLAVES
; i
++) {
984 if (i
== slp
- &slaves
[0]) {
990 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
991 (slaves
[i
].pid
= fork()) < 0)
992 quit("too many slaves, %d (recompile smaller): %s\n",
995 slaves
[i
].fd
= cmd
[1];
997 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
999 for (j
= 0; j
<= i
; j
++)
1000 (void) close(slaves
[j
].fd
);
1002 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1003 #if defined(SIGINFO)
1004 sigaddset(&sigs
, SIGINFO
);
1006 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1008 #ifdef LINUX_FORK_BUG
1009 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1011 quit("master/slave protocol botched 3\n");
1020 #ifdef LINUX_FORK_BUG
1022 * Wait for all slaves to _actually_ start to circumvent a bug in
1023 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1024 * returned from fork() causes a SEGV in the child process
1026 for (i
= 0; i
< SLAVES
; i
++)
1027 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1028 quit("master/slave protocol botched 4\n");
1031 for (i
= 0; i
< SLAVES
; i
++)
1032 (void) atomic_write( slaves
[i
].fd
,
1033 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1034 sizeof slaves
[0].pid
);
1044 for (i
= 0; i
< SLAVES
; i
++)
1045 if (slaves
[i
].pid
> 0) {
1046 (void) kill(slaves
[i
].pid
, SIGKILL
);
1052 * Synchronization - each process waits for a SIGUSR2 from the
1053 * previous process before writing to the tape, and sends SIGUSR2
1054 * to the next process when the tape write completes. On tape errors
1055 * a SIGUSR1 is sent to the master which then terminates all of the
1059 doslave(int cmd
, int slave_number
)
1062 int nextslave
, size
, eot_count
, bufsize
;
1063 volatile int wrote
= 0;
1066 struct tapebuf
*comp_buf
= NULL
;
1067 int compresult
, do_compress
= slave_number
> 0;
1068 unsigned long worklen
;
1069 #endif /* HAVE_ZLIB */
1070 struct slave_results returns
;
1076 union u_spcl
*uspclptr
;
1077 struct s_spcl
*spclptr
;
1078 #endif /* USE_QFA */
1081 * Need our own seek pointer.
1083 (void) close(diskfd
);
1084 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1085 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1088 retval
= dump_fs_open(disk
, &fs
);
1090 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1091 #endif /* __linux__ */
1094 * Need the pid of the next slave in the loop...
1096 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1097 != sizeof nextslave
) {
1098 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1102 /* if we're doing a compressed dump, allocate the compress buffer */
1104 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1105 if (comp_buf
== NULL
)
1106 quit("couldn't allocate a compress buffer.\n");
1107 comp_buf
->flags
= 0;
1109 #endif /* HAVE_ZLIB */
1112 * Get list of blocks to dump, read the blocks into tape buffer
1114 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1115 register struct req
*p
= slp
->req
;
1117 for (trecno
= 0; trecno
< ntrec
;
1118 trecno
+= p
->count
, p
+= p
->count
) {
1119 if (p
->dblk
) { /* read a disk block */
1120 bread(p
->dblk
, slp
->tblock
[trecno
],
1121 p
->count
* TP_BSIZE
);
1122 } else { /* read record from pipe */
1123 if (p
->count
!= 1 || atomic_read( cmd
,
1124 (char *)slp
->tblock
[trecno
],
1125 TP_BSIZE
) != TP_BSIZE
)
1126 quit("master/slave protocol botched.\n");
1130 /* Try to write the data... */
1134 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1135 bufsize
= writesize
; /* length to write */
1136 returns
.clen
= returns
.unclen
= bufsize
;
1140 * When writing a compressed dump, each block except
1141 * the first one on each tape is written
1142 * from struct tapebuf with an 4 byte prefix
1143 * followed by the data. This can be less than
1144 * writesize. Restore, on a short read, can compare the
1145 * length read to the compressed length in the header
1146 * to verify that the read was good. Blocks which don't
1147 * compress well are written uncompressed.
1148 * The first block written by each slave is not compressed
1149 * and does not have a prefix.
1152 if (compressed
&& do_compress
) {
1153 comp_buf
->length
= bufsize
;
1154 worklen
= TP_BSIZE
+ writesize
;
1155 compresult
= compress2(comp_buf
->buf
, &worklen
,
1156 (char *)slp
->tblock
[0], writesize
, compressed
);
1157 if (compresult
== Z_OK
&& 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 */
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 /* if an error occured previously don't
1195 if ((gtperr
= GetTapePos(&curtapepos
)) == 0) {
1197 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1199 sprintf(gTps
, "%ld\t%d\t%ld\n", (unsigned long)spclptr
->c_inumber
, tapeno
, curtapepos
);
1200 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1201 warn("error writing tapepos file.\n");
1207 #endif /* USE_QFA */
1209 while (eot_count
< 10 && size
< bufsize
) {
1212 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1215 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1217 printf("slave %d wrote %d\n", slave_number
, wrote
);
1227 if (size
!= bufsize
)
1228 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1229 slave_number
, size
, bufsize
);
1233 * Handle ENOSPC as an EOT condition.
1235 if (wrote
< 0 && errno
== ENOSPC
) {
1241 returns
.clen
= returns
.unclen
= 0;
1244 * pass errno back to master for special handling
1247 returns
.unclen
= -errno
;
1250 * pass size of data and size of write back to master
1251 * (for EOT handling)
1253 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1256 * Signal the next slave to go.
1258 (void) kill(nextslave
, SIGUSR2
);
1261 quit("error reading command pipe: %s\n", strerror(errno
));
1265 * Since a read from a pipe may not return all we asked for,
1266 * or a write may not write all we ask if we get a signal,
1267 * loop until the count is satisfied (or error).
1270 atomic_read(int fd
, void *buf
, size_t count
)
1272 int got
, need
= count
;
1275 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1277 } while (got
== -1 && errno
== EINTR
);
1278 return (got
< 0 ? got
: count
- need
);
1282 * Since a read from a pipe may not return all we asked for,
1283 * or a write may not write all we ask if we get a signal,
1284 * loop until the count is satisfied (or error).
1287 atomic_write(int fd
, const void *buf
, size_t count
)
1289 int got
, need
= count
;
1292 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1294 } while (got
== -1 && errno
== EINTR
);
1295 return (got
< 0 ? got
: count
- need
);
1301 * read the current tape position
1304 GetTapePos(long *pos
)
1309 if (ioctl(tapefd
, MTIOCPOS
, pos
) == -1) {
1311 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1317 #endif /* USE_QFA */