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 1.68 2002/05/21 15:48:46 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>
74 #ifdef HAVE_EXT2FS_EXT2_FS_H
75 #include <ext2fs/ext2_fs.h>
77 #include <linux/ext2_fs.h>
79 #include <ext2fs/ext2fs.h>
81 #include <bsdcompat.h>
83 #include <sys/vnode.h>
86 #include <ufs/inode.h>
88 #include <ufs/ufs/dinode.h>
89 #include <ufs/ffs/fs.h>
90 #endif /* __linux__ */
92 #include <protocols/dumprestore.h>
96 #endif /* HAVE_ZLIB */
100 #endif /* HAVE_BZLIB */
104 int writesize
; /* size of malloc()ed buffer for tape */
105 long lastspclrec
= -1; /* tape block number of last written header */
106 int trecno
= 0; /* next record to write in current block */
107 extern long blocksperfile
; /* number of blocks per output file */
108 long blocksthisvol
; /* number of blocks on current output file */
109 extern int ntrec
; /* blocking factor on tape */
110 extern int cartridge
;
114 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
115 static int magtapeout
; /* output is really a tape */
117 static ssize_t dump_atomic_read
__P((int, void *, size_t));
118 static ssize_t dump_atomic_write
__P((int, const void *, size_t));
119 static void doslave
__P((int, int, int));
120 static void enslave
__P((void));
121 static void flushtape
__P((void));
122 static void killall
__P((void));
123 static void rollforward
__P((void));
125 static int GetTapePos
__P((long long *));
126 static void MkTapeString
__P((struct s_spcl
*, long long));
130 * Concurrent dump mods (Caltech) - disk block reading and tape writing
131 * are exported to several slave processes. While one slave writes the
132 * tape, the others read disk blocks; they pass control of the tape in
133 * a ring via signals. The parent process traverses the filesystem and
134 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
135 * The following structure defines the instruction packets sent to slaves.
143 struct slave_results
{
144 ssize_t unclen
; /* uncompressed length */
145 ssize_t clen
; /* compressed length */
148 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
150 int tapea
; /* header number at start of this chunk */
151 int count
; /* count to next header (used for TS_TAPE */
153 int inode
; /* inode that we are currently dealing with */
154 int fd
; /* FD for this slave */
155 int pid
; /* PID for this slave */
156 int sent
; /* 1 == we've sent this slave requests */
157 int firstrec
; /* record number of this block */
158 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
159 struct req
*req
; /* buffer for requests */
163 char (*nextblock
)[TP_BSIZE
];
165 static time_t tstart_volume
; /* time of volume start */
166 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
168 int master
; /* pid of master, for sending error signals */
169 int tenths
; /* length of tape overhead per block written */
170 static int caught
; /* have we caught the signal to proceed? */
171 static int ready
; /* have we reached the lock point without having */
172 /* received the SIGUSR2 signal from the prev slave? */
173 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
174 /* SIGUSR2 arrives from the previous slave */
176 static int gtperr
= 0;
182 int pgoff
= getpagesize() - 1;
186 writesize
= ntrec
* TP_BSIZE
;
187 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
189 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
190 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
191 * repositioning after stopping, i.e, streaming mode, where the gap is
192 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
194 if (blocksperfile
== 0 && !unlimited
)
195 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
201 * Allocate tape buffer contiguous with the array of instruction
202 * packets, so flushtape() can write them together with one write().
203 * Align tape buffer on page boundary to speed up tape write().
205 for (i
= 0; i
<= SLAVES
; i
++) {
207 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
210 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
212 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
214 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
216 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
222 nextblock
= slp
->tblock
;
227 writerec(const void *dp
, int isspcl
)
230 slp
->req
[trecno
].dblk
= (daddr_t
)0;
231 slp
->req
[trecno
].count
= 1;
232 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
233 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
235 /* Need to write it to the archive file */
236 if (Afile
< 0 && isspcl
&& (spcl
.c_type
== TS_END
))
239 /* When we dump an inode which is not a directory,
240 * it means we ended the archive contents */
241 if (isspcl
&& (spcl
.c_type
== TS_INODE
) &&
242 ((spcl
.c_dinode
.di_mode
& S_IFMT
) != IFDIR
))
246 tmp
= *(union u_spcl
*)dp
;
247 /* Write the record, _uncompressed_ */
249 tmp
.s_spcl
.c_flags
&= ~DR_COMPRESSED
;
252 if (write(Afile
, &tmp
, TP_BSIZE
) != TP_BSIZE
)
253 msg("error writing archive file: %s\n",
260 lastspclrec
= spcl
.c_tapea
;
268 dumpblock(daddr_t blkno
, int size
)
270 int avail
, tpblks
, dblkno
;
272 dblkno
= fsbtodb(sblock
, blkno
);
273 tpblks
= size
>> tp_bshift
;
274 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
275 slp
->req
[trecno
].dblk
= dblkno
;
276 slp
->req
[trecno
].count
= avail
;
278 spcl
.c_tapea
+= avail
;
281 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
293 msg("write error on %s: %s\n", tape
, strerror(errnum
));
294 quit("Cannot recover\n");
297 msg("write error %d blocks into volume %d: %s\n",
298 blocksthisvol
, tapeno
, strerror(errnum
));
299 broadcast("DUMP WRITE ERROR!\n");
300 if (query("Do you want to rewrite this volume?")) {
301 msg("Closing this volume. Prepare to restart with new media;\n");
302 msg("this dump volume will be rewritten.\n");
308 if (query("Do you want to start the next tape?"))
317 quit("Broken pipe\n");
322 * Update xferrate stats
331 ttaken
= tnow
- tstart_volume
;
332 blocks
= spcl
.c_tapea
- tapea_volume
;
333 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
335 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
336 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
338 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
339 long txfrate
= volkb
/ ttaken
;
340 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
341 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
342 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
346 double rate
= .0005 + (double) blocks
/ (double) volkb
;
347 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
349 tapeno
, blocks
, volkb
, rate
);
356 mktimeest(time_t tnow
)
358 static char msgbuf
[128];
363 if (blockswritten
< 500)
365 if (blockswritten
> tapesize
)
366 tapesize
= blockswritten
;
367 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
368 / blockswritten
* tapesize
;
369 if (tnow
> tstart_volume
)
370 (void)snprintf(msgbuf
, sizeof(msgbuf
),
371 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
372 (blockswritten
* 100.0) / tapesize
,
373 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
374 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
376 (void)snprintf(msgbuf
, sizeof(msgbuf
),
377 "%3.2f%% done, finished in %d:%02d\n",
378 (blockswritten
* 100.0) / tapesize
,
379 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
387 * information message upon receipt of SIGINFO
390 statussig(int notused
)
392 int save_errno
= errno
;
395 buf
= mktimeest(time(NULL
));
397 write(STDERR_FILENO
, buf
, strlen(buf
));
407 struct slave_results returned
;
409 int siz
= (char *)nextblock
- (char *)slp
->req
;
411 slp
->req
[trecno
].count
= 0; /* Sentinel */
413 if (dump_atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
414 quit("error writing command pipe: %s\n", strerror(errno
));
415 slp
->sent
= 1; /* we sent a request, read the response later */
417 lastfirstrec
= slp
->firstrec
;
419 if (++slp
>= &slaves
[SLAVES
])
422 /* Read results back from next slave */
424 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
425 != sizeof returned
) {
426 perror(" DUMP: error reading command pipe in master");
429 got
= returned
.unclen
;
430 bytes_written
+= returned
.clen
;
431 if (returned
.unclen
== returned
.clen
)
435 /* Check for errors or end of tape */
437 /* Check for errors */
441 msg("End of tape detected\n");
444 * Drain the results, don't care what the values were.
445 * If we read them here then trewind won't...
447 for (i
= 0; i
< SLAVES
; i
++) {
448 if (slaves
[i
].sent
) {
449 if (dump_atomic_read( slaves
[i
].fd
,
450 (char *)&returned
, sizeof returned
)
451 != sizeof returned
) {
452 perror(" DUMP: error reading command pipe in master");
466 if (spcl
.c_type
!= TS_END
) {
467 for (i
= 0; i
< spcl
.c_count
; i
++)
468 if (spcl
.c_addr
[i
] != 0)
471 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
472 slp
->tapea
= spcl
.c_tapea
;
473 slp
->firstrec
= lastfirstrec
+ ntrec
;
475 nextblock
= slp
->tblock
;
477 asize
+= tenths
+ returned
.clen
/ density
;
478 blockswritten
+= ntrec
;
479 blocksthisvol
+= ntrec
;
480 if (!pipeout
&& !unlimited
) {
482 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= blocksperfile
* 1024
483 : blocksthisvol
>= blocksperfile
) {
488 else if (asize
> tsize
) {
501 struct slave_results returned
;
503 for (f
= 0; f
< SLAVES
; f
++) {
505 * Drain the results, but unlike EOT we DO (or should) care
506 * what the return values were, since if we detect EOT after
507 * we think we've written the last blocks to the tape anyway,
508 * we have to replay those blocks with rollforward.
510 * fixme: punt for now.
512 if (slaves
[f
].sent
) {
513 if (dump_atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
514 != sizeof returned
) {
515 perror(" DUMP: error reading command pipe in master");
518 got
= returned
.unclen
;
519 bytes_written
+= returned
.clen
;
520 if (returned
.unclen
== returned
.clen
)
528 msg("EOT detected in last 2 tape records!\n");
529 msg("Use a longer tape, decrease the size estimate\n");
530 quit("or use no size estimate at all.\n");
533 (void) close(slaves
[f
].fd
);
535 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
540 msg("Closing %s\n", tape
);
545 while (rmtopen(tape
, "O_RDONLY") < 0)
552 (void) close(tapefd
);
554 while ((f
= OPEN(tape
, O_RDONLY
)) < 0)
560 if (eot_script
&& spcl
.c_type
!= TS_END
) {
561 msg("Launching %s\n", eot_script
);
562 eot_code
= system_command(eot_script
, tape
, tapeno
);
564 if (eot_code
!= 0 && eot_code
!= 1) {
565 msg("Dump aborted by the end of tape script\n");
577 if (nexttape
|| Mflag
|| (eot_code
== 0) )
580 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
581 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
583 while (!query("Is the new volume mounted and ready to go?"))
584 if (query("Do you want to abort?")) {
593 struct req
*p
, *q
= NULL
, *prev
;
595 int i
, size
, savedtapea
, got
;
596 union u_spcl
*ntb
, *otb
;
597 struct slave_results returned
;
602 tslp
= &slaves
[SLAVES
];
603 ntb
= (union u_spcl
*)tslp
->tblock
[1];
606 * Each of the N slaves should have requests that need to
607 * be replayed on the next tape. Use the extra slave buffers
608 * (slaves[SLAVES]) to construct request lists to be sent to
609 * each slave in turn.
611 for (i
= 0; i
< SLAVES
; i
++) {
613 otb
= (union u_spcl
*)slp
->tblock
;
616 * For each request in the current slave, copy it to tslp.
620 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
623 *ntb
++ = *otb
++; /* copy the datablock also */
628 quit("rollforward: protocol botch");
640 nextblock
= tslp
->tblock
;
641 savedtapea
= spcl
.c_tapea
;
642 spcl
.c_tapea
= slp
->tapea
;
644 spcl
.c_tapea
= savedtapea
;
645 lastspclrec
= savedtapea
- 1;
647 size
= (char *)ntb
- (char *)q
;
648 if (dump_atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
649 perror(" DUMP: error writing command pipe");
654 lastfirstrec
= slp
->firstrec
;
656 if (++slp
>= &slaves
[SLAVES
])
661 if (prev
->dblk
!= 0) {
663 * If the last one was a disk block, make the
664 * first of this one be the last bit of that disk
667 q
->dblk
= prev
->dblk
+
668 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
669 ntb
= (union u_spcl
*)tslp
->tblock
;
672 * It wasn't a disk block. Copy the data to its
673 * new location in the buffer.
676 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
677 ntb
= (union u_spcl
*)tslp
->tblock
[1];
681 nextblock
= slp
->tblock
;
684 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
685 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
692 * Clear the first slaves' response. One hopes that it
693 * worked ok, otherwise the tape is much too short!
696 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
697 != sizeof returned
) {
698 perror(" DUMP: error reading command pipe in master");
701 got
= returned
.unclen
;
702 bytes_written
+= returned
.clen
;
703 if (returned
.clen
== returned
.unclen
)
711 quit("EOT detected at start of the tape!\n");
717 if (spcl
.c_type
!= TS_END
) {
718 for (i
= 0; i
< spcl
.c_count
; i
++)
719 if (spcl
.c_addr
[i
] != 0)
723 slp
->firstrec
= lastfirstrec
+ ntrec
;
724 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
726 asize
+= tenths
+ returned
.clen
/ density
;
727 blockswritten
+= ntrec
;
728 blocksthisvol
+= ntrec
;
733 * We implement taking and restoring checkpoints on the tape level.
734 * When each tape is opened, a new process is created by forking; this
735 * saves all of the necessary context in the parent. The child
736 * continues the dump; the parent waits around, saving the context.
737 * If the child returns X_REWRITE, then it had problems writing that tape;
738 * this causes the parent to fork again, duplicating the context, and
739 * everything continues as if nothing had happened.
742 startnewtape(int top
)
753 sigaddset(&sigs
, SIGINT
);
754 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
755 #else /* __linux__ */
757 void (*interrupt_save
)();
759 sig_t interrupt_save
;
761 interrupt_save
= signal(SIGINT
, SIG_IGN
);
762 #endif /* __linux__ */
764 parentpid
= getpid();
765 tapea_volume
= spcl
.c_tapea
;
766 tapea_bytes
= bytes_written
;
767 tstart_volume
= time(NULL
);
771 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
773 (void)signal(SIGINT
, interrupt_save
);
776 * All signals are inherited...
780 msg("Context save fork fails in parent %d\n", parentpid
);
786 * save the context by waiting
787 * until the child doing all of the work returns.
788 * don't catch the interrupt
791 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
793 signal(SIGINT
, SIG_IGN
);
796 msg("Tape: %d; parent process: %d child process %d\n",
797 tapeno
+1, parentpid
, childpid
);
799 while ((waitpid
= wait(&status
)) != childpid
)
800 if (waitpid
!= rshpid
)
801 msg("Parent %d waiting for child %d has another child %d return\n",
802 parentpid
, childpid
, waitpid
);
804 msg("Child %d returns LOB status %o\n",
805 childpid
, status
&0xFF);
807 status
= (status
>> 8) & 0xFF;
811 msg("Child %d finishes X_FINOK\n", childpid
);
814 msg("Child %d finishes X_ABORT\n", childpid
);
817 msg("Child %d finishes X_REWRITE\n", childpid
);
820 msg("Child %d finishes unknown %d\n",
831 goto restore_check_point
;
833 msg("Bad return code from dump: %d\n", status
);
837 } else { /* we are the child; just continue */
839 sleep(4); /* allow time for parent's message to get out */
840 msg("Child on Tape %d has parent %d, my pid = %d\n",
841 tapeno
+1, parentpid
, getpid());
844 * If we have a name like "/dev/rmt0,/dev/rmt1",
845 * use the name before the comma first, and save
846 * the remaining names for subsequent volumes.
848 tapeno
++; /* current tape sequence */
850 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
851 tape
[MAXPATHLEN
- 1] = '\0';
852 msg("Dumping volume %d on %s\n", tapeno
, tape
);
854 else if (nexttape
|| strchr(tapeprefix
, ',')) {
855 if (nexttape
&& *nexttape
)
856 tapeprefix
= nexttape
;
857 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
862 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
863 tape
[MAXPATHLEN
- 1] = '\0';
864 msg("Dumping volume %d on %s\n", tapeno
, tape
);
867 while ((tapefd
= (host
? rmtopen(tape
, "O_WRONLY|O_CREAT|O_TRUNC") : pipeout
?
869 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
871 while ((tapefd
= (pipeout
? fileno(stdout
) :
872 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
875 msg("Cannot open output \"%s\": %s\n", tape
,
877 if (!query("Do you want to retry the open?"))
884 struct mtget mt_stat
;
885 magtapeout
= ioctl(tapefd
, MTIOCGET
, (char *)&mt_stat
) == 0;
887 msg("Output is to %s\n",
888 magtapeout ? "tape" : "file/pipe");
892 enslave(); /* Share open tape file descriptor with slaves */
897 newtape
++; /* new tape signal */
898 spcl
.c_count
= slp
->count
;
900 * measure firstrec in TP_BSIZE units since restore doesn't
901 * know the correct ntrec value...
903 spcl
.c_firstrec
= slp
->firstrec
;
905 spcl
.c_type
= TS_TAPE
;
906 spcl
.c_flags
|= DR_NEWHEADER
;
907 spcl
.c_ntrec
= ntrec
;
909 spcl
.c_flags
|= DR_COMPRESSED
;
910 writeheader((dump_ino_t
)slp
->inode
);
911 spcl
.c_flags
&=~ DR_NEWHEADER
;
912 msg("Volume %d started with block %ld at: %s", tapeno
,
913 spcl
.c_tapea
, ctime(&tstart_volume
));
915 msg("Volume %d begins with blocks from inode %d\n",
917 if (tapeno
< TP_NINOS
)
918 volinfo
[tapeno
] = slp
->inode
;
926 if (master
!= 0 && master
!= getpid())
927 /* Signals master to call dumpabort */
928 (void) kill(master
, SIGTERM
);
931 msg("The ENTIRE dump is aborted.\n");
944 msg("pid = %d exits with status %d\n", getpid(), status
);
950 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
956 siglongjmp(jmpbuf
, 1);
964 #ifdef LINUX_FORK_BUG
972 { struct sigaction sa
;
973 memset(&sa
, 0, sizeof sa
);
974 sigemptyset(&sa
.sa_mask
);
975 sa
.sa_handler
= dumpabort
;
976 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
977 sa
.sa_handler
= sigpipe
;
978 sigaction(SIGPIPE
, &sa
, NULL
);
979 sa
.sa_handler
= proceed
;
980 sa
.sa_flags
= SA_RESTART
;
981 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
984 for (i
= 0; i
< SLAVES
; i
++) {
985 if (i
== slp
- &slaves
[0]) {
991 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
992 (slaves
[i
].pid
= fork()) < 0)
993 quit("too many slaves, %d (recompile smaller): %s\n",
996 slaves
[i
].fd
= cmd
[1];
998 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
1000 for (j
= 0; j
<= i
; j
++)
1001 (void) close(slaves
[j
].fd
);
1003 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1004 #if defined(SIGINFO)
1005 sigaddset(&sigs
, SIGINFO
);
1007 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1009 #ifdef LINUX_FORK_BUG
1010 if (dump_atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1012 quit("master/slave protocol botched 3\n");
1014 doslave(cmd
[0], i
, (slaves
[i
].pid
== slp
->pid
));
1021 #ifdef LINUX_FORK_BUG
1023 * Wait for all slaves to _actually_ start to circumvent a bug in
1024 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1025 * returned from fork() causes a SEGV in the child process
1027 for (i
= 0; i
< SLAVES
; i
++)
1028 if (dump_atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1029 quit("master/slave protocol botched 4\n");
1032 for (i
= 0; i
< SLAVES
; i
++)
1033 (void) dump_atomic_write( slaves
[i
].fd
,
1034 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1035 sizeof slaves
[0].pid
);
1045 for (i
= 0; i
< SLAVES
; i
++)
1046 if (slaves
[i
].pid
> 0) {
1047 (void) kill(slaves
[i
].pid
, SIGKILL
);
1053 * Synchronization - each process waits for a SIGUSR2 from the
1054 * previous process before writing to the tape, and sends SIGUSR2
1055 * to the next process when the tape write completes. On tape errors
1056 * a SIGUSR1 is sent to the master which then terminates all of the
1060 doslave(int cmd
, int slave_number
, int first
)
1063 int nextslave
, size
, eot_count
, bufsize
;
1064 volatile int wrote
= 0;
1066 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1067 struct tapebuf
*comp_buf
= NULL
;
1068 int compresult
, do_compress
= !first
;
1069 unsigned long worklen
;
1071 unsigned int worklen2
;
1073 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1074 struct slave_results returns
;
1079 long long curtapepos
;
1080 union u_spcl
*uspclptr
;
1081 struct s_spcl
*spclptr
;
1082 #endif /* USE_QFA */
1086 sigaddset(&set
, SIGUSR2
);
1087 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1091 * Need our own seek pointer.
1093 (void) close(diskfd
);
1094 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1095 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1098 (void)ioctl(diskfd
, BLKFLSBUF
);
1101 retval
= dump_fs_open(disk
, &fs
);
1103 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1104 #endif /* __linux__ */
1107 * Need the pid of the next slave in the loop...
1109 if ((nread
= dump_atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1110 != sizeof nextslave
) {
1111 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1114 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1115 /* if we're doing a compressed dump, allocate the compress buffer */
1117 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1118 if (comp_buf
== NULL
)
1119 quit("couldn't allocate a compress buffer.\n");
1121 comp_buf
->flags
= COMPRESS_BZLIB
;
1123 comp_buf
->flags
= COMPRESS_ZLIB
;
1125 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1128 * Get list of blocks to dump, read the blocks into tape buffer
1130 while ((nread
= dump_atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1131 struct req
*p
= slp
->req
;
1133 for (trecno
= 0; trecno
< ntrec
;
1134 trecno
+= p
->count
, p
+= p
->count
) {
1135 if (p
->dblk
) { /* read a disk block */
1136 bread(p
->dblk
, slp
->tblock
[trecno
],
1137 p
->count
* TP_BSIZE
);
1138 } else { /* read record from pipe */
1139 if (p
->count
!= 1 || dump_atomic_read( cmd
,
1140 (char *)slp
->tblock
[trecno
],
1141 TP_BSIZE
) != TP_BSIZE
)
1142 quit("master/slave protocol botched.\n");
1146 /* Try to write the data... */
1150 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1151 bufsize
= writesize
; /* length to write */
1152 returns
.clen
= returns
.unclen
= bufsize
;
1154 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1156 * When writing a compressed dump, each block except
1157 * the first one on each tape is written
1158 * from struct tapebuf with an 4 byte prefix
1159 * followed by the data. This can be less than
1160 * writesize. Restore, on a short read, can compare the
1161 * length read to the compressed length in the header
1162 * to verify that the read was good. Blocks which don't
1163 * compress well are written uncompressed.
1164 * The first block written by each slave is not compressed
1165 * and does not have a prefix.
1168 if (compressed
&& do_compress
) {
1169 comp_buf
->length
= bufsize
;
1170 worklen
= TP_BSIZE
+ writesize
;
1174 compresult
= compress2(comp_buf
->buf
,
1176 (char *)slp
->tblock
[0],
1179 if (compresult
== Z_OK
)
1184 #endif /* HAVE_ZLIB */
1188 compresult
= BZ2_bzBuffToBuffCompress(
1191 (char *)slp
->tblock
[0],
1196 if (compresult
== BZ_OK
)
1202 #endif /* HAVE_BZLIB */
1203 if (compresult
&& worklen
<= (writesize
- 16)) {
1204 /* write the compressed buffer */
1205 comp_buf
->length
= worklen
;
1206 comp_buf
->compressed
= 1;
1207 buffer
= (char *) comp_buf
;
1208 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1211 /* write the data uncompressed */
1212 comp_buf
->length
= writesize
;
1213 comp_buf
->compressed
= 0;
1214 buffer
= (char *) comp_buf
;
1215 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1216 returns
.unclen
= returns
.clen
;
1217 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1220 /* compress the remaining blocks if we're compressing */
1221 do_compress
= compressed
;
1222 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1224 if (sigsetjmp(jmpbuf
, 1) == 0) {
1233 if (gTapeposfd
>= 0) {
1236 for (i
= 0; i
< ntrec
; ++i
) {
1237 uspclptr
= (union u_spcl
*)&slp
->tblock
[i
];
1238 spclptr
= &uspclptr
->s_spcl
;
1239 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1240 (spclptr
->c_type
== TS_INODE
) &&
1241 ((spclptr
->c_dinode
.di_mode
& S_IFMT
) != IFDIR
) &&
1242 (spclptr
->c_date
== gThisDumpDate
)) {
1243 /* if an error occured previously don't
1248 gtperr
= GetTapePos(&curtapepos
);
1251 MkTapeString(spclptr
, curtapepos
);
1255 #endif /* USE_QFA */
1257 while (eot_count
< 10 && size
< bufsize
) {
1260 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1263 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1265 printf("slave %d wrote %d\n", slave_number
, wrote
);
1275 if (size
!= bufsize
)
1276 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1277 slave_number
, size
, bufsize
);
1281 * Handle ENOSPC as an EOT condition.
1283 if (wrote
< 0 && errno
== ENOSPC
) {
1289 returns
.clen
= returns
.unclen
= 0;
1292 * pass errno back to master for special handling
1295 returns
.unclen
= -errno
;
1298 * pass size of data and size of write back to master
1299 * (for EOT handling)
1301 (void) dump_atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1304 * Signal the next slave to go.
1306 (void) kill(nextslave
, SIGUSR2
);
1309 quit("error reading command pipe: %s\n", strerror(errno
));
1313 * Since a read from a pipe may not return all we asked for,
1314 * or a write may not write all we ask if we get a signal,
1315 * loop until the count is satisfied (or error).
1318 dump_atomic_read(int fd
, void *buf
, size_t count
)
1320 int got
, need
= count
;
1323 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1325 } while (got
== -1 && errno
== EINTR
);
1326 return (got
< 0 ? got
: count
- need
);
1330 * Since a read from a pipe may not return all we asked for,
1331 * or a write may not write all we ask if we get a signal,
1332 * loop until the count is satisfied (or error).
1335 dump_atomic_write(int fd
, const void *buf
, size_t count
)
1337 int got
, need
= count
;
1340 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1342 } while (got
== -1 && errno
== EINTR
);
1343 return (got
< 0 ? got
: count
- need
);
1349 * read the current tape position
1352 GetTapePos(long long *pos
)
1358 *pos
= (long long) rmtseek(0, SEEK_CUR
);
1367 err
= (ioctl(tapefd
, MTIOCPOS
, &mtpos
) < 0);
1368 *pos
= (long long)mtpos
;
1371 *pos
= LSEEK(tapefd
, 0, SEEK_CUR
);
1377 msg("[%ld] error: %d (getting tapepos: %lld)\n", getpid(),
1385 MkTapeString(struct s_spcl
*spclptr
, long long curtapepos
) {
1388 msg("inode %ld at tapepos %lld\n", spclptr
->c_inumber
, curtapepos
);
1391 snprintf(gTps
, sizeof(gTps
), "%ld\t%d\t%lld\n",
1392 (unsigned long)spclptr
->c_inumber
,
1395 gTps
[sizeof(gTps
) - 1] = '\0';
1396 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1397 warn("error writing tapepos file.\n");
1400 #endif /* USE_QFA */