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. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 static const char rcsid
[] =
40 "$Id: tape.c,v 1.81 2004/03/01 10:52:53 stelian Exp $";
44 #include <compatlfs.h>
50 #include <compaterr.h>
61 #include <sys/types.h>
63 #include <linux/fs.h> /* for definition of BLKFLSBUF */
66 #include <sys/param.h>
67 #include <sys/socket.h>
71 #ifdef HAVE_EXT2FS_EXT2_FS_H
72 #include <ext2fs/ext2_fs.h>
74 #include <linux/ext2_fs.h>
76 #include <ext2fs/ext2fs.h>
78 #include <bsdcompat.h>
80 #include <sys/vnode.h>
83 #include <ufs/inode.h>
85 #include <ufs/ufs/dinode.h>
86 #include <ufs/ffs/fs.h>
87 #endif /* __linux__ */
89 #include <protocols/dumprestore.h>
93 #endif /* HAVE_ZLIB */
97 #endif /* HAVE_BZLIB */
101 #endif /* HAVE_LZO */
105 int writesize
; /* size of malloc()ed buffer for tape */
106 long lastspclrec
= -1; /* tape block number of last written header */
107 int trecno
= 0; /* next record to write in current block */
108 extern long *blocksperfiles
; /* number of blocks per output file(s) */
109 long blocksperfiles_current
; /* current position in blocksperfiles */
110 long blocksthisvol
; /* number of blocks on current output file */
111 extern int ntrec
; /* blocking factor on tape */
112 extern int cartridge
;
116 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
117 static int magtapeout
; /* output is really a tape */
119 static ssize_t dump_atomic_read
__P((int, void *, size_t));
120 static ssize_t dump_atomic_write
__P((int, const void *, size_t));
122 static void doslave
__P((int, int, int));
124 static void doslave
__P((int, int));
126 static void enslave
__P((void));
127 static void flushtape
__P((void));
128 static void killall
__P((void));
129 static void rollforward
__P((void));
131 static int GetTapePos
__P((long long *));
132 static int MkTapeString
__P((struct s_spcl
*, long long));
133 #define FILESQFAPOS 20
137 * Concurrent dump mods (Caltech) - disk block reading and tape writing
138 * are exported to several slave processes. While one slave writes the
139 * tape, the others read disk blocks; they pass control of the tape in
140 * a ring via signals. The parent process traverses the filesystem and
141 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
142 * The following structure defines the instruction packets sent to slaves.
150 struct slave_results
{
151 ssize_t unclen
; /* uncompressed length */
152 ssize_t clen
; /* compressed length */
155 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
157 int tapea
; /* header number at start of this chunk */
158 int count
; /* count to next header (used for TS_TAPE */
160 int inode
; /* inode that we are currently dealing with */
161 int fd
; /* FD for this slave */
162 int pid
; /* PID for this slave */
163 int sent
; /* 1 == we've sent this slave requests */
164 int firstrec
; /* record number of this block */
165 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
166 struct req
*req
; /* buffer for requests */
170 char (*nextblock
)[TP_BSIZE
];
172 static time_t tstart_volume
; /* time of volume start */
173 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
175 int master
; /* pid of master, for sending error signals */
176 int tenths
; /* length of tape overhead per block written */
177 static int caught
; /* have we caught the signal to proceed? */
178 static int ready
; /* have we reached the lock point without having */
179 /* received the SIGUSR2 signal from the prev slave? */
180 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
181 /* SIGUSR2 arrives from the previous slave */
183 static int gtperr
= 0;
189 int pgoff
= getpagesize() - 1;
193 writesize
= ntrec
* TP_BSIZE
;
194 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
196 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
197 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
198 * repositioning after stopping, i.e, streaming mode, where the gap is
199 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
201 if (!blocksperfiles
&& !unlimited
)
202 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
208 * Allocate tape buffer contiguous with the array of instruction
209 * packets, so flushtape() can write them together with one write().
210 * Align tape buffer on page boundary to speed up tape write().
212 for (i
= 0; i
<= SLAVES
; i
++) {
214 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
217 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
219 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
221 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
223 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
229 nextblock
= slp
->tblock
;
234 writerec(const void *dp
, int isspcl
)
237 slp
->req
[trecno
].dblk
= (daddr_t
)0;
238 slp
->req
[trecno
].count
= 1;
239 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
240 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
242 /* Need to write it to the archive file */
243 if (! AfileActive
&& isspcl
&& (spcl
.c_type
== TS_END
))
245 if (AfileActive
&& Afile
>= 0) {
246 /* When we dump an inode which is not a directory,
247 * it means we ended the archive contents */
248 if (isspcl
&& (spcl
.c_type
== TS_INODE
) &&
249 ((spcl
.c_dinode
.di_mode
& S_IFMT
) != IFDIR
))
253 tmp
= *(union u_spcl
*)dp
;
254 /* Write the record, _uncompressed_ */
256 tmp
.s_spcl
.c_flags
&= ~DR_COMPRESSED
;
259 if (write(Afile
, &tmp
, TP_BSIZE
) != TP_BSIZE
)
260 msg("error writing archive file: %s\n",
267 lastspclrec
= spcl
.c_tapea
;
275 dumpblock(daddr_t blkno
, int size
)
277 int avail
, tpblks
, dblkno
;
279 dblkno
= fsbtodb(sblock
, blkno
);
280 tpblks
= size
>> tp_bshift
;
281 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
282 slp
->req
[trecno
].dblk
= dblkno
;
283 slp
->req
[trecno
].count
= avail
;
285 spcl
.c_tapea
+= avail
;
288 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
300 msg("write error on %s: %s\n", tape
, strerror(errnum
));
301 quit("Cannot recover\n");
304 msg("write error %d blocks into volume %d: %s\n",
305 blocksthisvol
, tapeno
, strerror(errnum
));
306 broadcast("DUMP WRITE ERROR!\n");
307 if (query("Do you want to rewrite this volume?")) {
308 msg("Closing this volume. Prepare to restart with new media;\n");
309 msg("this dump volume will be rewritten.\n");
315 if (query("Do you want to start the next tape?"))
321 sigpipe(UNUSED(int signo
))
324 quit("Broken pipe\n");
329 * Update xferrate stats
338 ttaken
= tnow
- tstart_volume
;
339 blocks
= spcl
.c_tapea
- tapea_volume
;
340 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
342 msg("Volume %d %ld blocks (%.2fMB)\n", tapeno
,
343 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
345 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
346 long txfrate
= volkb
/ ttaken
;
347 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
348 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
349 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
353 double rate
= .0005 + (double) blocks
/ (double) volkb
;
354 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
356 tapeno
, blocks
, volkb
, rate
);
363 mktimeest(time_t tnow
)
365 static char msgbuf
[128];
370 if (blockswritten
< 500)
372 if (blockswritten
> tapesize
)
373 tapesize
= blockswritten
;
374 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
375 / blockswritten
* tapesize
;
376 if (tnow
> tstart_volume
)
377 (void)snprintf(msgbuf
, sizeof(msgbuf
),
378 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
379 (blockswritten
* 100.0) / tapesize
,
380 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
381 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
383 (void)snprintf(msgbuf
, sizeof(msgbuf
),
384 "%3.2f%% done, finished in %d:%02d\n",
385 (blockswritten
* 100.0) / tapesize
,
386 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
394 * information message upon receipt of SIGINFO
397 statussig(int notused
)
399 int save_errno
= errno
;
402 buf
= mktimeest(time(NULL
));
404 write(STDERR_FILENO
, buf
, strlen(buf
));
414 struct slave_results returned
;
416 int siz
= (char *)nextblock
- (char *)slp
->req
;
418 slp
->req
[trecno
].count
= 0; /* Sentinel */
420 if (dump_atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
421 quit("error writing command pipe: %s\n", strerror(errno
));
422 slp
->sent
= 1; /* we sent a request, read the response later */
424 lastfirstrec
= slp
->firstrec
;
426 if (++slp
>= &slaves
[SLAVES
])
429 /* Read results back from next slave */
431 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
432 != sizeof returned
) {
433 perror(" DUMP: error reading command pipe in master");
436 got
= returned
.unclen
;
437 bytes_written
+= returned
.clen
;
438 if (returned
.unclen
== returned
.clen
)
442 /* Check for errors or end of tape */
444 /* Check for errors */
448 msg("End of tape detected\n");
451 * Drain the results, don't care what the values were.
452 * If we read them here then trewind won't...
454 for (i
= 0; i
< SLAVES
; i
++) {
455 if (slaves
[i
].sent
) {
456 if (dump_atomic_read( slaves
[i
].fd
,
457 (char *)&returned
, sizeof returned
)
458 != sizeof returned
) {
459 perror(" DUMP: error reading command pipe in master");
473 if (spcl
.c_type
!= TS_END
) {
474 for (i
= 0; i
< spcl
.c_count
; i
++)
475 if (spcl
.c_addr
[i
] != 0)
478 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
479 slp
->tapea
= spcl
.c_tapea
;
480 slp
->firstrec
= lastfirstrec
+ ntrec
;
482 nextblock
= slp
->tblock
;
484 asize
+= tenths
+ returned
.clen
/ density
;
485 blockswritten
+= ntrec
;
486 blocksthisvol
+= ntrec
;
487 if (!pipeout
&& !unlimited
) {
488 if (blocksperfiles
&& blocksperfiles
[blocksperfiles_current
]) {
489 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= (((long long)blocksperfiles
[blocksperfiles_current
]) * 1024)
490 : blocksthisvol
>= blocksperfiles
[blocksperfiles_current
] ) {
495 else if (asize
> tsize
) {
508 struct slave_results returned
;
510 for (f
= 0; f
< SLAVES
; f
++) {
512 * Drain the results, but unlike EOT we DO (or should) care
513 * what the return values were, since if we detect EOT after
514 * we think we've written the last blocks to the tape anyway,
515 * we have to replay those blocks with rollforward.
517 * fixme: punt for now.
519 if (slaves
[f
].sent
) {
520 if (dump_atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
521 != sizeof returned
) {
522 perror(" DUMP: error reading command pipe in master");
525 got
= returned
.unclen
;
526 bytes_written
+= returned
.clen
;
527 if (returned
.unclen
== returned
.clen
)
535 msg("EOT detected in last 2 tape records!\n");
536 msg("Use a longer tape, decrease the size estimate\n");
537 quit("or use no size estimate at all.\n");
540 (void) close(slaves
[f
].fd
);
542 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
547 msg("Closing %s\n", tape
);
552 while (rmtopen(tape
, O_RDONLY
) < 0)
559 (void) close(tapefd
);
561 while ((f
= OPEN(tape
, O_RDONLY
)) < 0)
576 if (nexttape
|| Mflag
)
579 msg("Launching %s\n", eot_script
);
580 eot_code
= system_command(eot_script
, tape
, tapeno
);
582 if (eot_code
!= 0 && eot_code
!= 1) {
583 msg("Dump aborted by the end of tape script\n");
589 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
590 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
592 while (!query("Is the new volume mounted and ready to go?"))
593 if (query("Do you want to abort?")) {
602 struct req
*p
, *q
= NULL
, *prev
;
604 int i
, size
, savedtapea
, got
;
605 union u_spcl
*ntb
, *otb
;
606 struct slave_results returned
;
611 tslp
= &slaves
[SLAVES
];
612 ntb
= (union u_spcl
*)tslp
->tblock
[1];
615 * Each of the N slaves should have requests that need to
616 * be replayed on the next tape. Use the extra slave buffers
617 * (slaves[SLAVES]) to construct request lists to be sent to
618 * each slave in turn.
620 for (i
= 0; i
< SLAVES
; i
++) {
622 otb
= (union u_spcl
*)slp
->tblock
;
625 * For each request in the current slave, copy it to tslp.
629 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
632 *ntb
++ = *otb
++; /* copy the datablock also */
637 quit("rollforward: protocol botch");
649 nextblock
= tslp
->tblock
;
650 savedtapea
= spcl
.c_tapea
;
651 spcl
.c_tapea
= slp
->tapea
;
653 spcl
.c_tapea
= savedtapea
;
654 lastspclrec
= savedtapea
- 1;
656 size
= (char *)ntb
- (char *)q
;
657 if (dump_atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
658 perror(" DUMP: error writing command pipe");
663 lastfirstrec
= slp
->firstrec
;
665 if (++slp
>= &slaves
[SLAVES
])
670 if (prev
->dblk
!= 0) {
672 * If the last one was a disk block, make the
673 * first of this one be the last bit of that disk
676 q
->dblk
= prev
->dblk
+
677 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
678 ntb
= (union u_spcl
*)tslp
->tblock
;
681 * It wasn't a disk block. Copy the data to its
682 * new location in the buffer.
685 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
686 ntb
= (union u_spcl
*)tslp
->tblock
[1];
690 nextblock
= slp
->tblock
;
693 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
694 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
701 * Clear the first slaves' response. One hopes that it
702 * worked ok, otherwise the tape is much too short!
705 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
706 != sizeof returned
) {
707 perror(" DUMP: error reading command pipe in master");
710 got
= returned
.unclen
;
711 bytes_written
+= returned
.clen
;
712 if (returned
.clen
== returned
.unclen
)
720 quit("EOT detected at start of the tape!\n");
726 if (spcl
.c_type
!= TS_END
) {
727 for (i
= 0; i
< spcl
.c_count
; i
++)
728 if (spcl
.c_addr
[i
] != 0)
732 slp
->firstrec
= lastfirstrec
+ ntrec
;
733 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
735 asize
+= tenths
+ returned
.clen
/ density
;
736 blockswritten
+= ntrec
;
737 blocksthisvol
+= ntrec
;
742 * We implement taking and restoring checkpoints on the tape level.
743 * When each tape is opened, a new process is created by forking; this
744 * saves all of the necessary context in the parent. The child
745 * continues the dump; the parent waits around, saving the context.
746 * If the child returns X_REWRITE, then it had problems writing that tape;
747 * this causes the parent to fork again, duplicating the context, and
748 * everything continues as if nothing had happened.
751 startnewtape(int top
)
762 sigaddset(&sigs
, SIGINT
);
763 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
764 #else /* __linux__ */
766 void (*interrupt_save
)();
768 sig_t interrupt_save
;
770 interrupt_save
= signal(SIGINT
, SIG_IGN
);
771 #endif /* __linux__ */
773 parentpid
= getpid();
774 tapea_volume
= spcl
.c_tapea
;
775 tapea_bytes
= bytes_written
;
776 tstart_volume
= time(NULL
);
780 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
782 (void)signal(SIGINT
, interrupt_save
);
785 * All signals are inherited...
789 msg("Context save fork fails in parent %d\n", parentpid
);
795 * save the context by waiting
796 * until the child doing all of the work returns.
797 * don't catch the interrupt
800 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
802 signal(SIGINT
, SIG_IGN
);
805 msg("Tape: %d; parent process: %d child process %d\n",
806 tapeno
+1, parentpid
, childpid
);
808 while ((waitpid
= wait(&status
)) != childpid
)
809 if (waitpid
!= rshpid
)
810 msg("Parent %d waiting for child %d has another child %d return\n",
811 parentpid
, childpid
, waitpid
);
813 msg("Child %d returns LOB status %o\n",
814 childpid
, status
&0xFF);
816 status
= (status
>> 8) & 0xFF;
820 msg("Child %d finishes X_FINOK\n", childpid
);
823 msg("Child %d finishes X_ABORT\n", childpid
);
826 msg("Child %d finishes X_REWRITE\n", childpid
);
829 msg("Child %d finishes unknown %d\n",
840 goto restore_check_point
;
842 msg("Bad return code from dump: %d\n", status
);
846 } else { /* we are the child; just continue */
848 sleep(4); /* allow time for parent's message to get out */
849 msg("Child on Tape %d has parent %d, my pid = %d\n",
850 tapeno
+1, parentpid
, getpid());
853 * If we have a name like "/dev/rmt0,/dev/rmt1",
854 * use the name before the comma first, and save
855 * the remaining names for subsequent volumes.
857 tapeno
++; /* current tape sequence */
859 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
860 tape
[MAXPATHLEN
- 1] = '\0';
861 msg("Dumping volume %d on %s\n", tapeno
, tape
);
863 else if (nexttape
|| strchr(tapeprefix
, ',')) {
864 if (nexttape
&& *nexttape
)
865 tapeprefix
= nexttape
;
866 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
871 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
872 tape
[MAXPATHLEN
- 1] = '\0';
873 msg("Dumping volume %d on %s\n", tapeno
, tape
);
875 if (blocksperfiles
&& blocksperfiles_current
< *blocksperfiles
)
876 blocksperfiles_current
++;
878 while ((tapefd
= (host
? rmtopen(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
) : pipeout
?
880 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
882 while ((tapefd
= (pipeout
? fileno(stdout
) :
883 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
886 msg("Cannot open output \"%s\": %s\n", tape
,
888 if (!query("Do you want to retry the open?"))
895 struct mtget mt_stat
;
896 magtapeout
= ioctl(tapefd
, MTIOCGET
, (char *)&mt_stat
) == 0;
898 msg("Output is to %s\n",
899 magtapeout ? "tape" : "file/pipe");
903 enslave(); /* Share open tape file descriptor with slaves */
908 newtape
++; /* new tape signal */
909 spcl
.c_count
= slp
->count
;
911 * measure firstrec in TP_BSIZE units since restore doesn't
912 * know the correct ntrec value...
914 spcl
.c_firstrec
= slp
->firstrec
;
916 spcl
.c_type
= TS_TAPE
;
917 spcl
.c_flags
|= DR_NEWHEADER
;
918 spcl
.c_ntrec
= ntrec
;
920 spcl
.c_flags
|= DR_COMPRESSED
;
921 writeheader((dump_ino_t
)slp
->inode
);
922 spcl
.c_flags
&=~ DR_NEWHEADER
;
923 msg("Volume %d started with block %ld at: %s", tapeno
,
924 spcl
.c_tapea
, ctime(&tstart_volume
));
926 msg("Volume %d begins with blocks from inode %d\n",
928 if (tapeno
< (int)TP_NINOS
)
929 volinfo
[tapeno
] = slp
->inode
;
934 dumpabort(UNUSED(int signo
))
937 if (master
!= 0 && master
!= getpid())
938 /* Signals master to call dumpabort */
939 (void) kill(master
, SIGTERM
);
942 msg("The ENTIRE dump is aborted.\n");
955 msg("pid = %d exits with status %d\n", getpid(), status
);
961 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
964 proceed(UNUSED(int signo
))
967 siglongjmp(jmpbuf
, 1);
975 #ifdef LINUX_FORK_BUG
983 { struct sigaction sa
;
984 memset(&sa
, 0, sizeof sa
);
985 sigemptyset(&sa
.sa_mask
);
986 sa
.sa_handler
= dumpabort
;
987 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
988 sa
.sa_handler
= sigpipe
;
989 sigaction(SIGPIPE
, &sa
, NULL
);
990 sa
.sa_handler
= proceed
;
991 sa
.sa_flags
= SA_RESTART
;
992 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
995 for (i
= 0; i
< SLAVES
; i
++) {
996 if (i
== slp
- &slaves
[0]) {
1002 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
1003 (slaves
[i
].pid
= fork()) < 0)
1004 quit("too many slaves, %d (recompile smaller): %s\n",
1005 i
, strerror(errno
));
1007 slaves
[i
].fd
= cmd
[1];
1009 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
1011 for (j
= 0; j
<= i
; j
++)
1012 (void) close(slaves
[j
].fd
);
1014 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1015 #if defined(SIGINFO)
1016 sigaddset(&sigs
, SIGINFO
);
1018 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1020 #ifdef LINUX_FORK_BUG
1021 if (dump_atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1023 quit("master/slave protocol botched 3\n");
1029 (slaves
[i
].pid
== slp
->pid
));
1036 #ifdef LINUX_FORK_BUG
1038 * Wait for all slaves to _actually_ start to circumvent a bug in
1039 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1040 * returned from fork() causes a SEGV in the child process
1042 for (i
= 0; i
< SLAVES
; i
++)
1043 if (dump_atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1044 quit("master/slave protocol botched 4\n");
1047 for (i
= 0; i
< SLAVES
; i
++)
1048 (void) dump_atomic_write( slaves
[i
].fd
,
1049 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1050 sizeof slaves
[0].pid
);
1060 for (i
= 0; i
< SLAVES
; i
++)
1061 if (slaves
[i
].pid
> 0) {
1062 (void) kill(slaves
[i
].pid
, SIGKILL
);
1068 * Synchronization - each process waits for a SIGUSR2 from the
1069 * previous process before writing to the tape, and sends SIGUSR2
1070 * to the next process when the tape write completes. On tape errors
1071 * a SIGUSR1 is sent to the master which then terminates all of the
1083 volatile int wrote
= 0, size
, eot_count
, bufsize
;
1084 char * volatile buffer
;
1085 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1086 struct tapebuf
* volatile comp_buf
= NULL
;
1088 volatile int do_compress
= !first
;
1089 unsigned long worklen
;
1091 unsigned int worklen2
;
1094 lzo_align_t __LZO_MMODEL
*LZO_WorkMem
;
1096 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1097 struct slave_results returns
;
1102 long long curtapepos
;
1103 union u_spcl
*uspclptr
;
1104 struct s_spcl
*spclptr
;
1105 /* long maxntrecs = 300000000 / (ntrec * 1024); last tested: 50 000 000 */
1106 long maxntrecs
= 50000; /* every 50MB */
1107 long cntntrecs
= maxntrecs
;
1108 #endif /* USE_QFA */
1112 sigaddset(&set
, SIGUSR2
);
1113 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1117 * Need our own seek pointer.
1119 (void) close(diskfd
);
1120 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1121 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1124 (void)ioctl(diskfd
, BLKFLSBUF
, 0);
1127 retval
= dump_fs_open(disk
, &fs
);
1129 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1130 #endif /* __linux__ */
1133 * Need the pid of the next slave in the loop...
1135 if ((nread
= dump_atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1136 != sizeof nextslave
) {
1137 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1140 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1141 /* if we're doing a compressed dump, allocate the compress buffer */
1143 int bsiz
= sizeof(struct tapebuf
) + writesize
;
1144 /* Add extra space to deal with compression enlarging the buffer */
1145 if (TP_BSIZE
> writesize
/16 + 67)
1148 bsiz
+= writesize
/16 + 67;
1149 comp_buf
= malloc(bsiz
);
1150 if (comp_buf
== NULL
)
1151 quit("couldn't allocate a compress buffer.\n");
1152 if (zipflag
== COMPRESS_ZLIB
)
1153 comp_buf
->flags
= COMPRESS_ZLIB
;
1154 else if (zipflag
== COMPRESS_BZLIB
)
1155 comp_buf
->flags
= COMPRESS_BZLIB
;
1156 else if (zipflag
== COMPRESS_LZO
) {
1157 comp_buf
->flags
= COMPRESS_LZO
;
1158 if (lzo_init() != LZO_E_OK
) quit("lzo_init failed\n");
1160 quit("internal error - unknown compression method: %d\n", zipflag
);
1163 LZO_WorkMem
= malloc(LZO1X_1_MEM_COMPRESS
);
1165 quit("couldn't allocate a compress buffer.\n");
1167 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1170 * Get list of blocks to dump, read the blocks into tape buffer
1172 while ((nread
= dump_atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1173 struct req
*p
= slp
->req
;
1175 for (trecno
= 0; trecno
< ntrec
;
1176 trecno
+= p
->count
, p
+= p
->count
) {
1177 if (p
->dblk
) { /* read a disk block */
1178 bread(p
->dblk
, slp
->tblock
[trecno
],
1179 p
->count
* TP_BSIZE
);
1180 } else { /* read record from pipe */
1181 if (p
->count
!= 1 || dump_atomic_read( cmd
,
1182 (char *)slp
->tblock
[trecno
],
1183 TP_BSIZE
) != TP_BSIZE
)
1184 quit("master/slave protocol botched.\n");
1188 /* Try to write the data... */
1192 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1193 bufsize
= writesize
; /* length to write */
1194 returns
.clen
= returns
.unclen
= bufsize
;
1196 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1198 * When writing a compressed dump, each block except
1199 * the first one on each tape is written
1200 * from struct tapebuf with an 4 byte prefix
1201 * followed by the data. This can be less than
1202 * writesize. Restore, on a short read, can compare the
1203 * length read to the compressed length in the header
1204 * to verify that the read was good. Blocks which don't
1205 * compress well are written uncompressed.
1206 * The first block written by each slave is not compressed
1207 * and does not have a prefix.
1210 if (compressed
&& do_compress
) {
1211 comp_buf
->length
= bufsize
;
1212 worklen
= TP_BSIZE
+ writesize
;
1215 if (zipflag
== COMPRESS_ZLIB
) {
1216 compresult
= compress2(comp_buf
->buf
,
1218 (char *)slp
->tblock
[0],
1221 if (compresult
== Z_OK
)
1226 #endif /* HAVE_ZLIB */
1228 if (zipflag
== COMPRESS_BZLIB
) {
1230 compresult
= BZ2_bzBuffToBuffCompress(
1233 (char *)slp
->tblock
[0],
1238 if (compresult
== BZ_OK
)
1244 #endif /* HAVE_BZLIB */
1246 if (zipflag
== COMPRESS_LZO
) {
1247 compresult
= lzo1x_1_compress((char *)slp
->tblock
[0],writesize
,
1249 (lzo_uintp
)&worklen
,
1251 if (compresult
== LZO_E_OK
)
1256 #endif /* HAVE_LZO */
1257 if (compresult
&& worklen
<= ((unsigned long)writesize
- 16)) {
1258 /* write the compressed buffer */
1259 comp_buf
->length
= worklen
;
1260 comp_buf
->compressed
= 1;
1261 buffer
= (char *) comp_buf
;
1262 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1265 /* write the data uncompressed */
1266 comp_buf
->length
= writesize
;
1267 comp_buf
->compressed
= 0;
1268 buffer
= (char *) comp_buf
;
1269 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1270 returns
.unclen
= returns
.clen
;
1271 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1274 /* compress the remaining blocks if we're compressing */
1275 do_compress
= compressed
;
1276 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1278 if (sigsetjmp(jmpbuf
, 1) == 0) {
1287 if (gTapeposfd
>= 0) {
1291 for (i
= 0; (i
< ntrec
) && !foundone
; ++i
) {
1292 uspclptr
= (union u_spcl
*)&slp
->tblock
[i
];
1293 spclptr
= &uspclptr
->s_spcl
;
1294 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1295 (spclptr
->c_type
== TS_INODE
) &&
1296 (spclptr
->c_date
== gThisDumpDate
) &&
1297 !(spclptr
->c_dinode
.di_mode
& S_IFDIR
)
1300 /* if (cntntrecs >= maxntrecs) { only write every maxntrecs amount of data */
1303 gtperr
= GetTapePos(&curtapepos
);
1304 /* if an error occured previously don't
1308 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1310 gtperr
= MkTapeString(spclptr
, curtapepos
);
1316 #endif /* USE_QFA */
1318 while (eot_count
< 10 && size
< bufsize
) {
1321 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1324 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1326 printf("slave %d wrote %d\n", slave_number
, wrote
);
1328 if (wrote
< 0 && errno
!= ENOSPC
)
1330 if (wrote
== 0 || (wrote
< 0 && errno
== ENOSPC
))
1337 if (size
!= bufsize
)
1338 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1339 slave_number
, size
, bufsize
);
1343 * Handle ENOSPC as an EOT condition.
1345 if (wrote
< 0 && errno
== ENOSPC
) {
1351 returns
.clen
= returns
.unclen
= 0;
1354 * pass errno back to master for special handling
1357 returns
.unclen
= -errno
;
1360 * pass size of data and size of write back to master
1361 * (for EOT handling)
1363 (void) dump_atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1366 * Signal the next slave to go.
1368 (void) kill(nextslave
, SIGUSR2
);
1370 if (gTapeposfd
>= 0) {
1373 #endif /* USE_QFA */
1376 quit("error reading command pipe: %s\n", strerror(errno
));
1380 * Since a read from a pipe may not return all we asked for,
1381 * or a write may not write all we ask if we get a signal,
1382 * loop until the count is satisfied (or error).
1385 dump_atomic_read(int fd
, void *buf
, size_t count
)
1387 int got
, need
= count
;
1390 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1392 } while (got
== -1 && errno
== EINTR
);
1393 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1397 * Since a read from a pipe may not return all we asked for,
1398 * or a write may not write all we ask if we get a signal,
1399 * loop until the count is satisfied (or error).
1402 dump_atomic_write(int fd
, const void *buf
, size_t count
)
1404 int got
, need
= count
;
1407 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1409 } while (got
== -1 && errno
== EINTR
);
1410 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1421 buf.mt_op = MTSETDRVBUFFER;
1422 buf.mt_count = MT_ST_BOOLEANS | MT_ST_SCSI2LOGICAL;
1423 if (ioctl(tapefd, MTIOCTOP, &buf) == -1) {
1425 msg("[%ld] error: %d (setting logical)\n",
1426 (unsigned long)getpid(), err);
1433 #define LSEEK_GET_TAPEPOS 10
1434 #define LSEEK_GO2_TAPEPOS 11
1436 * read the current tape position
1439 GetTapePos(long long *pos
)
1445 *pos
= (long long) rmtseek((OFF_T
)0, (int)LSEEK_GET_TAPEPOS
);
1454 err
= (ioctl(tapefd
, MTIOCPOS
, &mtpos
) < 0);
1455 *pos
= (long long)mtpos
;
1458 *pos
= LSEEK(tapefd
, 0, SEEK_CUR
);
1464 msg("[%ld] error: %d (getting tapepos: %lld)\n", getpid(),
1472 MkTapeString(struct s_spcl
*spclptr
, long long curtapepos
)
1477 msg("inode %ld at tapepos %lld\n", spclptr
->c_inumber
, curtapepos
);
1480 snprintf(gTps
, sizeof(gTps
), "%ld\t%d\t%lld\n",
1481 (unsigned long)spclptr
->c_inumber
,
1484 gTps
[sizeof(gTps
) - 1] = '\0';
1485 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != (ssize_t
)strlen(gTps
)) {
1487 warn("error writing tapepos file. (error %d)\n", errno
);
1491 #endif /* USE_QFA */