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 1.92 2011/05/20 09:48:40 stelian Exp $";
44 #include <compatlfs.h>
50 #include <compaterr.h>
61 #include <sys/types.h>
63 #include <sys/ioctl.h>
64 #include <sys/mount.h> /* for definition of BLKFLSBUF */
65 #ifndef BLKFLSBUF /* last resort... */
66 #define BLKFLSBUF _IO(0x12, 97) /* Flush buffer cache. */
70 #include <sys/param.h>
71 #include <sys/socket.h>
75 #ifdef HAVE_EXT2FS_EXT2_FS_H
76 #include <ext2fs/ext2_fs.h>
78 #include <linux/ext2_fs.h>
80 #include <ext2fs/ext2fs.h>
82 #include <bsdcompat.h>
84 #include <sys/vnode.h>
87 #include <ufs/inode.h>
89 #include <ufs/ufs/dinode.h>
90 #include <ufs/ffs/fs.h>
91 #endif /* __linux__ */
93 #include <protocols/dumprestore.h>
97 #endif /* HAVE_ZLIB */
101 #endif /* HAVE_BZLIB */
105 #endif /* HAVE_LZO */
109 int writesize
; /* size of malloc()ed buffer for tape */
110 long lastspclrec
= -1; /* tape block number of last written header */
111 int trecno
= 0; /* next record to write in current block */
112 extern long *blocksperfiles
; /* number of blocks per output file(s) */
113 long blocksperfiles_current
; /* current position in blocksperfiles */
114 long blocksthisvol
; /* number of blocks on current output file */
115 extern int ntrec
; /* blocking factor on tape */
116 extern int cartridge
;
120 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
121 static int magtapeout
; /* output is really a tape */
123 static ssize_t dump_atomic_read
__P((int, char *, size_t));
124 static ssize_t dump_atomic_write
__P((int, const char *, size_t));
126 static void doslave
__P((int, int, int));
128 static void doslave
__P((int, int));
130 static void enslave
__P((void));
131 static void flushtape
__P((void));
132 static void killall
__P((void));
133 static void rollforward
__P((void));
135 static int GetTapePos
__P((long long *));
136 static int MkTapeString
__P((struct s_spcl
*, long long));
137 #define FILESQFAPOS 20
141 * Concurrent dump mods (Caltech) - disk block reading and tape writing
142 * are exported to several slave processes. While one slave writes the
143 * tape, the others read disk blocks; they pass control of the tape in
144 * a ring via signals. The parent process traverses the filesystem and
145 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
146 * The following structure defines the instruction packets sent to slaves.
154 struct slave_results
{
155 ssize_t unclen
; /* uncompressed length */
156 ssize_t clen
; /* compressed length */
159 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
161 int tapea
; /* header number at start of this chunk */
162 int count
; /* count to next header (used for TS_TAPE */
164 int inode
; /* inode that we are currently dealing with */
165 int fd
; /* FD for this slave */
166 int pid
; /* PID for this slave */
167 int sent
; /* 1 == we've sent this slave requests */
168 int firstrec
; /* record number of this block */
169 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
170 struct req
*req
; /* buffer for requests */
174 char (*nextblock
)[TP_BSIZE
];
176 static time_t tstart_volume
; /* time of volume start */
177 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
179 int master
; /* pid of master, for sending error signals */
180 int tenths
; /* length of tape overhead per block written */
181 static int caught
; /* have we caught the signal to proceed? */
182 static int ready
; /* have we reached the lock point without having */
183 /* received the SIGUSR2 signal from the prev slave? */
184 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
185 /* SIGUSR2 arrives from the previous slave */
187 static int gtperr
= 0;
191 * Determine if we can use Linux' clone system call. If so, call it
192 * with the CLONE_IO flag so that all processes will share the same I/O
193 * context, allowing the I/O schedulers to make better scheduling decisions.
196 /* first, pull in the header files that define sys_clone and CLONE_IO */
203 /* If either is not present, fall back on the fork behaviour */
204 #if ! defined(SYS_clone) || ! defined (CLONE_IO)
205 #define fork_clone_io fork
206 #else /* SYS_clone */
207 /* CLONE_IO is available, determine which version of sys_clone to use */
208 #include <linux/version.h>
210 * Kernel 2.5.49 introduced two extra parameters to the clone system call.
211 * Neither is useful in our case, so this is easy to handle.
213 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,49)
215 /* child_stack, clone_flags, parent_tidptr, child_tidptr */
216 #define CLONE_ARGS 0, SIGCHLD|CLONE_IO, NULL, NULL
218 /* clone_flags, child_stack, parent_tidptr, child_tidptr */
219 #define CLONE_ARGS SIGCHLD|CLONE_IO, 0, NULL, NULL
220 #endif /* __s390__ */
222 #define CLONE_ARGS SIGCHLD|CLONE_IO, 0
223 #endif /* LINUX_VERSION_CODE */
224 pid_t
fork_clone_io(void);
225 #endif /* SYS_clone */
226 #else /* __linux__ not defined */
227 #define fork_clone_io fork
228 #endif /* __linux__ */
233 int pgoff
= getpagesize() - 1;
237 writesize
= ntrec
* TP_BSIZE
;
238 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
240 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
241 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
242 * repositioning after stopping, i.e, streaming mode, where the gap is
243 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
245 if (!blocksperfiles
&& !unlimited
)
246 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
252 * Allocate tape buffer contiguous with the array of instruction
253 * packets, so flushtape() can write them together with one write().
254 * Align tape buffer on page boundary to speed up tape write().
256 for (i
= 0; i
<= SLAVES
; i
++) {
258 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
261 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
263 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
265 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
267 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
273 nextblock
= slp
->tblock
;
278 writerec(const void *dp
, int isspcl
)
281 slp
->req
[trecno
].dblk
= (ext2_loff_t
)0;
282 slp
->req
[trecno
].count
= 1;
283 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
284 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
286 /* Need to write it to the archive file */
287 if (! AfileActive
&& isspcl
&& (spcl
.c_type
== TS_END
))
289 if (AfileActive
&& Afile
>= 0 && !(spcl
.c_flags
& DR_EXTATTRIBUTES
)) {
290 /* When we dump an inode which is not a directory,
291 * it means we ended the archive contents */
292 if (isspcl
&& (spcl
.c_type
== TS_INODE
) &&
293 ((spcl
.c_dinode
.di_mode
& S_IFMT
) != IFDIR
))
297 tmp
= *(union u_spcl
*)dp
;
298 /* Write the record, _uncompressed_ */
300 tmp
.s_spcl
.c_flags
&= ~DR_COMPRESSED
;
303 if (write(Afile
, &tmp
, TP_BSIZE
) != TP_BSIZE
)
304 msg("error writing archive file: %s\n",
311 lastspclrec
= spcl
.c_tapea
;
319 dumpblock(blk_t blkno
, int size
)
324 dblkno
= fsbtodb(sblock
, blkno
);
325 tpblks
= size
>> tp_bshift
;
326 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
327 slp
->req
[trecno
].dblk
= dblkno
;
328 slp
->req
[trecno
].count
= avail
;
330 spcl
.c_tapea
+= avail
;
333 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
345 msg("write error on %s: %s\n", tape
, strerror(errnum
));
346 quit("Cannot recover\n");
349 msg("write error %d blocks into volume %d: %s\n",
350 blocksthisvol
, tapeno
, strerror(errnum
));
351 broadcast("DUMP WRITE ERROR!\n");
352 if (query("Do you want to rewrite this volume?")) {
353 msg("Closing this volume. Prepare to restart with new media;\n");
354 msg("this dump volume will be rewritten.\n");
360 if (query("Do you want to start the next tape?"))
366 sigpipe(UNUSED(int signo
))
369 quit("Broken pipe\n");
374 * Update xferrate stats
383 ttaken
= tnow
- tstart_volume
;
384 blocks
= spcl
.c_tapea
- tapea_volume
;
385 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
387 msg("Volume %d %ld blocks (%.2fMB)\n", tapeno
,
388 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
390 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
391 long txfrate
= volkb
/ ttaken
;
392 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
393 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
394 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
398 double rate
= .0005 + (double) blocks
/ (double) volkb
;
399 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
401 tapeno
, blocks
, volkb
, rate
);
408 mktimeest(time_t tnow
)
410 static char msgbuf
[128];
415 if (blockswritten
< 500)
417 if (blockswritten
> tapesize
)
418 tapesize
= blockswritten
;
419 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
420 / blockswritten
* tapesize
;
421 if (tnow
> tstart_volume
)
422 (void)snprintf(msgbuf
, sizeof(msgbuf
),
423 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
424 (blockswritten
* 100.0) / tapesize
,
425 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
426 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
428 (void)snprintf(msgbuf
, sizeof(msgbuf
),
429 "%3.2f%% done, finished in %d:%02d\n",
430 (blockswritten
* 100.0) / tapesize
,
431 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
439 * information message upon receipt of SIGINFO
442 statussig(int notused
)
444 int save_errno
= errno
;
447 buf
= mktimeest(time(NULL
));
449 write(STDERR_FILENO
, buf
, strlen(buf
));
459 struct slave_results returned
;
461 int siz
= (char *)nextblock
- (char *)slp
->req
;
463 /* make sure returned has sane values in case we don't read
464 * them from the slave in this pass */
465 returned
.unclen
= returned
.clen
= writesize
;
467 slp
->req
[trecno
].count
= 0; /* Sentinel */
469 if (dump_atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
470 quit("error writing command pipe: %s\n", strerror(errno
));
471 slp
->sent
= 1; /* we sent a request, read the response later */
473 lastfirstrec
= slp
->firstrec
;
475 if (++slp
>= &slaves
[SLAVES
])
478 /* Read results back from next slave */
480 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
481 != sizeof returned
) {
482 perror(" DUMP: error reading command pipe in master");
485 got
= returned
.unclen
;
486 bytes_written
+= returned
.clen
;
487 if (returned
.unclen
== returned
.clen
)
491 /* Check for errors or end of tape */
493 /* Check for errors */
497 msg("End of tape detected\n");
500 * Drain the results, don't care what the values were.
501 * If we read them here then trewind won't...
503 for (i
= 0; i
< SLAVES
; i
++) {
504 if (slaves
[i
].sent
) {
505 if (dump_atomic_read( slaves
[i
].fd
,
506 (char *)&returned
, sizeof returned
)
507 != sizeof returned
) {
508 perror(" DUMP: error reading command pipe in master");
522 if (spcl
.c_type
== TS_CLRI
|| spcl
.c_type
== TS_BITS
)
525 if (spcl
.c_type
!= TS_END
) {
526 for (i
= 0; i
< spcl
.c_count
; i
++)
527 if (spcl
.c_addr
[i
] != 0)
531 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
532 slp
->tapea
= spcl
.c_tapea
;
533 slp
->firstrec
= lastfirstrec
+ ntrec
;
535 nextblock
= slp
->tblock
;
537 asize
+= tenths
+ returned
.clen
/ density
;
538 blockswritten
+= ntrec
;
539 blocksthisvol
+= ntrec
;
540 if (!pipeout
&& !unlimited
) {
541 if (blocksperfiles
&& blocksperfiles
[blocksperfiles_current
]) {
542 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= (((long long)blocksperfiles
[blocksperfiles_current
]) * 1024)
543 : blocksthisvol
>= blocksperfiles
[blocksperfiles_current
] ) {
548 else if (asize
> tsize
) {
561 struct slave_results returned
;
563 for (f
= 0; f
< SLAVES
; f
++) {
565 * Drain the results, but unlike EOT we DO (or should) care
566 * what the return values were, since if we detect EOT after
567 * we think we've written the last blocks to the tape anyway,
568 * we have to replay those blocks with rollforward.
570 * fixme: punt for now.
572 if (slaves
[f
].sent
) {
573 if (dump_atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
574 != sizeof returned
) {
575 perror(" DUMP: error reading command pipe in master");
578 got
= returned
.unclen
;
579 bytes_written
+= returned
.clen
;
580 if (returned
.unclen
== returned
.clen
)
588 msg("EOT detected in last 2 tape records!\n");
589 msg("Use a longer tape, decrease the size estimate\n");
590 quit("or use no size estimate at all.\n");
593 (void) close(slaves
[f
].fd
);
595 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
600 msg("Closing %s\n", tape
);
605 while (rmtopen(tape
, O_RDONLY
) < 0)
612 (void) close(tapefd
);
614 while ((f
= OPEN(tape
, O_RDONLY
)) < 0)
630 msg("Launching %s\n", eot_script
);
631 eot_code
= system_command(eot_script
, tape
, tapeno
);
633 if (eot_code
!= 0 && eot_code
!= 1) {
634 msg("Dump aborted by the end of tape script\n");
639 if (nexttape
|| Mflag
)
642 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
643 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
645 while (!query("Is the new volume mounted and ready to go?"))
646 if (query("Do you want to abort?")) {
655 struct req
*p
, *q
= NULL
, *prev
;
657 int i
, size
, savedtapea
, got
;
658 union u_spcl
*ntb
, *otb
;
659 struct slave_results returned
;
664 tslp
= &slaves
[SLAVES
];
665 ntb
= (union u_spcl
*)tslp
->tblock
[1];
667 /* make sure returned has sane values in case we don't read
668 * them from the slave in this pass */
669 returned
.unclen
= returned
.clen
= writesize
;
672 * Each of the N slaves should have requests that need to
673 * be replayed on the next tape. Use the extra slave buffers
674 * (slaves[SLAVES]) to construct request lists to be sent to
675 * each slave in turn.
677 for (i
= 0; i
< SLAVES
; i
++) {
679 otb
= (union u_spcl
*)slp
->tblock
;
682 * For each request in the current slave, copy it to tslp.
686 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
689 *ntb
++ = *otb
++; /* copy the datablock also */
694 quit("rollforward: protocol botch");
706 nextblock
= tslp
->tblock
;
707 savedtapea
= spcl
.c_tapea
;
708 spcl
.c_tapea
= slp
->tapea
;
710 spcl
.c_tapea
= savedtapea
;
711 lastspclrec
= savedtapea
- 1;
713 size
= (char *)ntb
- (char *)q
;
714 if (dump_atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
715 perror(" DUMP: error writing command pipe");
720 lastfirstrec
= slp
->firstrec
;
722 if (++slp
>= &slaves
[SLAVES
])
727 if (prev
->dblk
!= 0) {
729 * If the last one was a disk block, make the
730 * first of this one be the last bit of that disk
733 q
->dblk
= prev
->dblk
+
734 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
735 ntb
= (union u_spcl
*)tslp
->tblock
;
738 * It wasn't a disk block. Copy the data to its
739 * new location in the buffer.
742 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
743 ntb
= (union u_spcl
*)tslp
->tblock
[1];
747 nextblock
= slp
->tblock
;
750 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
751 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
758 * Clear the first slaves' response. One hopes that it
759 * worked ok, otherwise the tape is much too short!
762 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
763 != sizeof returned
) {
764 perror(" DUMP: error reading command pipe in master");
767 got
= returned
.unclen
;
768 bytes_written
+= returned
.clen
;
769 if (returned
.clen
== returned
.unclen
)
777 quit("EOT detected at start of the tape!\n");
783 if (spcl
.c_type
!= TS_END
) {
784 for (i
= 0; i
< spcl
.c_count
; i
++)
785 if (spcl
.c_addr
[i
] != 0)
789 slp
->firstrec
= lastfirstrec
+ ntrec
;
790 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
792 asize
+= tenths
+ returned
.clen
/ density
;
793 blockswritten
+= ntrec
;
794 blocksthisvol
+= ntrec
;
799 #if defined(SYS_clone) && defined(CLONE_IO)
803 return syscall(SYS_clone
, CLONE_ARGS
);
809 * We implement taking and restoring checkpoints on the tape level.
810 * When each tape is opened, a new process is created by forking; this
811 * saves all of the necessary context in the parent. The child
812 * continues the dump; the parent waits around, saving the context.
813 * If the child returns X_REWRITE, then it had problems writing that tape;
814 * this causes the parent to fork again, duplicating the context, and
815 * everything continues as if nothing had happened.
818 startnewtape(int top
)
829 sigaddset(&sigs
, SIGINT
);
830 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
831 #else /* __linux__ */
833 void (*interrupt_save
)();
835 sig_t interrupt_save
;
837 interrupt_save
= signal(SIGINT
, SIG_IGN
);
838 #endif /* __linux__ */
840 parentpid
= getpid();
841 tapea_volume
= spcl
.c_tapea
;
842 tapea_bytes
= bytes_written
;
843 tstart_volume
= time(NULL
);
847 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
849 (void)signal(SIGINT
, interrupt_save
);
852 * All signals are inherited...
854 childpid
= fork_clone_io();
856 msg("Context save fork fails in parent %d\n", parentpid
);
862 * save the context by waiting
863 * until the child doing all of the work returns.
864 * don't catch the interrupt
867 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
869 signal(SIGINT
, SIG_IGN
);
872 msg("Tape: %d; parent process: %d child process %d\n",
873 tapeno
+1, parentpid
, childpid
);
875 while ((waitpid
= wait(&status
)) != childpid
)
876 if (waitpid
!= rshpid
)
877 msg("Parent %d waiting for child %d has another child %d return\n",
878 parentpid
, childpid
, waitpid
);
880 msg("Child %d returns LOB status %o\n",
881 childpid
, status
&0xFF);
883 status
= (status
>> 8) & 0xFF;
887 msg("Child %d finishes X_FINOK\n", childpid
);
890 msg("Child %d finishes X_ABORT\n", childpid
);
893 msg("Child %d finishes X_REWRITE\n", childpid
);
896 msg("Child %d finishes unknown %d\n",
907 goto restore_check_point
;
909 msg("Bad return code from dump: %d\n", status
);
913 } else { /* we are the child; just continue */
915 sleep(4); /* allow time for parent's message to get out */
916 msg("Child on Tape %d has parent %d, my pid = %d\n",
917 tapeno
+1, parentpid
, getpid());
920 * If we have a name like "/dev/rmt0,/dev/rmt1",
921 * use the name before the comma first, and save
922 * the remaining names for subsequent volumes.
924 tapeno
++; /* current tape sequence */
926 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
927 tape
[MAXPATHLEN
- 1] = '\0';
928 msg("Dumping volume %d on %s\n", tapeno
, tape
);
930 else if (nexttape
|| strchr(tapeprefix
, ',')) {
931 if (nexttape
&& *nexttape
)
932 tapeprefix
= nexttape
;
933 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
938 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
939 tape
[MAXPATHLEN
- 1] = '\0';
940 msg("Dumping volume %d on %s\n", tapeno
, tape
);
942 if (blocksperfiles
&& blocksperfiles_current
< *blocksperfiles
)
943 blocksperfiles_current
++;
945 while ((tapefd
= (host
? rmtopen(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
) : pipeout
?
947 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
949 while ((tapefd
= (pipeout
? fileno(stdout
) :
950 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
953 msg("Cannot open output \"%s\": %s\n", tape
,
955 if (!query("Do you want to retry the open?"))
962 struct mtget mt_stat
;
963 magtapeout
= ioctl(tapefd
, MTIOCGET
, (char *)&mt_stat
) == 0;
965 msg("Output is to %s\n",
966 magtapeout ? "tape" : "file/pipe");
970 enslave(); /* Share open tape file descriptor with slaves */
975 newtape
++; /* new tape signal */
976 spcl
.c_count
= slp
->count
;
978 * measure firstrec in TP_BSIZE units since restore doesn't
979 * know the correct ntrec value...
981 spcl
.c_firstrec
= slp
->firstrec
;
983 spcl
.c_type
= TS_TAPE
;
984 spcl
.c_flags
|= DR_NEWHEADER
;
985 spcl
.c_ntrec
= ntrec
;
987 spcl
.c_flags
|= DR_COMPRESSED
;
988 writeheader((dump_ino_t
)slp
->inode
);
989 spcl
.c_flags
&=~ DR_NEWHEADER
;
990 msg("Volume %d started with block %ld at: %s", tapeno
,
991 spcl
.c_tapea
, ctime(&tstart_volume
));
993 msg("Volume %d begins with blocks from inode %d\n",
995 if (tapeno
< (int)TP_NINOS
)
996 volinfo
[tapeno
] = slp
->inode
;
1001 dumpabort(UNUSED(int signo
))
1004 if (master
!= 0 && master
!= getpid())
1005 /* Signals master to call dumpabort */
1006 (void) kill(master
, SIGTERM
);
1009 msg("The ENTIRE dump is aborted.\n");
1022 msg("pid = %d exits with status %d\n", getpid(), status
);
1028 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
1031 proceed(UNUSED(int signo
))
1034 siglongjmp(jmpbuf
, 1);
1042 #ifdef LINUX_FORK_BUG
1050 { struct sigaction sa
;
1051 memset(&sa
, 0, sizeof sa
);
1052 sigemptyset(&sa
.sa_mask
);
1053 sa
.sa_handler
= dumpabort
;
1054 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
1055 sa
.sa_handler
= sigpipe
;
1056 sigaction(SIGPIPE
, &sa
, NULL
);
1057 sa
.sa_handler
= proceed
;
1058 sa
.sa_flags
= SA_RESTART
;
1059 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
1062 for (i
= 0; i
< SLAVES
; i
++) {
1063 if (i
== slp
- &slaves
[0]) {
1069 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
1070 (slaves
[i
].pid
= fork_clone_io()) < 0)
1071 quit("too many slaves, %d (recompile smaller): %s\n",
1072 i
, strerror(errno
));
1074 slaves
[i
].fd
= cmd
[1];
1076 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
1078 for (j
= 0; j
<= i
; j
++)
1079 (void) close(slaves
[j
].fd
);
1081 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1082 #if defined(SIGINFO)
1083 sigaddset(&sigs
, SIGINFO
);
1085 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1087 #ifdef LINUX_FORK_BUG
1088 if (dump_atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1090 quit("master/slave protocol botched 3\n");
1096 (slaves
[i
].pid
== slp
->pid
));
1103 #ifdef LINUX_FORK_BUG
1105 * Wait for all slaves to _actually_ start to circumvent a bug in
1106 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1107 * returned from fork() causes a SEGV in the child process
1109 for (i
= 0; i
< SLAVES
; i
++)
1110 if (dump_atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1111 quit("master/slave protocol botched 4\n");
1114 for (i
= 0; i
< SLAVES
; i
++)
1115 (void) dump_atomic_write( slaves
[i
].fd
,
1116 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1117 sizeof slaves
[0].pid
);
1127 for (i
= 0; i
< SLAVES
; i
++)
1128 if (slaves
[i
].pid
> 0) {
1129 (void) kill(slaves
[i
].pid
, SIGKILL
);
1135 * Synchronization - each process waits for a SIGUSR2 from the
1136 * previous process before writing to the tape, and sends SIGUSR2
1137 * to the next process when the tape write completes. On tape errors
1138 * a SIGUSR1 is sent to the master which then terminates all of the
1150 volatile int wrote
= 0, size
, eot_count
, bufsize
;
1151 char * volatile buffer
;
1152 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1153 struct tapebuf
* volatile comp_buf
= NULL
;
1155 volatile int do_compress
= !first
;
1156 unsigned long worklen
;
1158 lzo_align_t __LZO_MMODEL
*LZO_WorkMem
;
1160 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1161 struct slave_results returns
;
1166 long long curtapepos
;
1167 union u_spcl
*uspclptr
;
1168 struct s_spcl
*spclptr
;
1169 /* long maxntrecs = 300000000 / (ntrec * 1024); last tested: 50 000 000 */
1170 long maxntrecs
= 50000; /* every 50MB */
1171 long cntntrecs
= maxntrecs
;
1172 #endif /* USE_QFA */
1176 sigaddset(&set
, SIGUSR2
);
1177 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1181 * Need our own seek pointer.
1183 (void) close(diskfd
);
1184 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1185 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1188 (void)ioctl(diskfd
, BLKFLSBUF
, 0);
1191 retval
= dump_fs_open(disk
, &fs
);
1193 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1194 #endif /* __linux__ */
1197 * Need the pid of the next slave in the loop...
1199 if ((nread
= dump_atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1200 != sizeof nextslave
) {
1201 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1204 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1205 /* if we're doing a compressed dump, allocate the compress buffer */
1207 int bsiz
= sizeof(struct tapebuf
) + writesize
;
1208 /* Add extra space to deal with compression enlarging the buffer */
1209 if (TP_BSIZE
> writesize
/16 + 67)
1212 bsiz
+= writesize
/16 + 67;
1213 comp_buf
= malloc(bsiz
);
1214 if (comp_buf
== NULL
)
1215 quit("couldn't allocate a compress buffer.\n");
1216 if (zipflag
== COMPRESS_ZLIB
)
1217 comp_buf
->flags
= COMPRESS_ZLIB
;
1218 else if (zipflag
== COMPRESS_BZLIB
)
1219 comp_buf
->flags
= COMPRESS_BZLIB
;
1220 else if (zipflag
== COMPRESS_LZO
) {
1221 comp_buf
->flags
= COMPRESS_LZO
;
1222 if (lzo_init() != LZO_E_OK
) quit("lzo_init failed\n");
1224 quit("internal error - unknown compression method: %d\n", zipflag
);
1227 LZO_WorkMem
= malloc(LZO1X_1_MEM_COMPRESS
);
1229 quit("couldn't allocate a compress buffer.\n");
1231 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1234 * Get list of blocks to dump, read the blocks into tape buffer
1236 while ((nread
= dump_atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1237 struct req
*p
= slp
->req
;
1239 for (trecno
= 0; trecno
< ntrec
;
1240 trecno
+= p
->count
, p
+= p
->count
) {
1241 if (p
->dblk
) { /* read a disk block */
1242 bread(p
->dblk
, slp
->tblock
[trecno
],
1243 p
->count
* TP_BSIZE
);
1244 } else { /* read record from pipe */
1245 if (p
->count
!= 1 || dump_atomic_read( cmd
,
1246 (char *)slp
->tblock
[trecno
],
1247 TP_BSIZE
) != TP_BSIZE
)
1248 quit("master/slave protocol botched.\n");
1252 /* Try to write the data... */
1256 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1257 bufsize
= writesize
; /* length to write */
1258 returns
.clen
= returns
.unclen
= bufsize
;
1260 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1262 * When writing a compressed dump, each block except
1263 * the first one on each tape is written
1264 * from struct tapebuf with an 4 byte prefix
1265 * followed by the data. This can be less than
1266 * writesize. Restore, on a short read, can compare the
1267 * length read to the compressed length in the header
1268 * to verify that the read was good. Blocks which don't
1269 * compress well are written uncompressed.
1270 * The first block written by each slave is not compressed
1271 * and does not have a prefix.
1274 if (compressed
&& do_compress
) {
1275 comp_buf
->length
= bufsize
;
1276 worklen
= TP_BSIZE
+ writesize
;
1279 if (zipflag
== COMPRESS_ZLIB
) {
1280 compresult
= compress2(comp_buf
->buf
,
1282 (char *)slp
->tblock
[0],
1285 if (compresult
== Z_OK
)
1290 #endif /* HAVE_ZLIB */
1292 if (zipflag
== COMPRESS_BZLIB
) {
1293 unsigned int worklen2
= worklen
;
1294 compresult
= BZ2_bzBuffToBuffCompress(
1297 (char *)slp
->tblock
[0],
1302 if (compresult
== BZ_OK
)
1308 #endif /* HAVE_BZLIB */
1310 if (zipflag
== COMPRESS_LZO
) {
1311 lzo_uint worklen2
= worklen
;
1312 compresult
= lzo1x_1_compress((char *)slp
->tblock
[0],writesize
,
1317 if (compresult
== LZO_E_OK
)
1322 #endif /* HAVE_LZO */
1323 if (compresult
&& worklen
<= ((unsigned long)writesize
- 16)) {
1324 /* write the compressed buffer */
1325 comp_buf
->length
= worklen
;
1326 comp_buf
->compressed
= 1;
1327 buffer
= (char *) comp_buf
;
1328 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1331 /* write the data uncompressed */
1332 comp_buf
->length
= writesize
;
1333 comp_buf
->compressed
= 0;
1334 buffer
= (char *) comp_buf
;
1335 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1336 returns
.unclen
= returns
.clen
;
1337 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1340 /* compress the remaining blocks if we're compressing */
1341 do_compress
= compressed
;
1342 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1344 if (sigsetjmp(jmpbuf
, 1) == 0) {
1353 if (gTapeposfd
>= 0) {
1357 for (i
= 0; (i
< ntrec
) && !foundone
; ++i
) {
1358 uspclptr
= (union u_spcl
*)&slp
->tblock
[i
];
1359 spclptr
= &uspclptr
->s_spcl
;
1360 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1361 (spclptr
->c_type
== TS_INODE
) &&
1362 (spclptr
->c_date
== gThisDumpDate
) &&
1363 !(spclptr
->c_dinode
.di_mode
& S_IFDIR
) &&
1364 !(spclptr
->c_flags
& DR_EXTATTRIBUTES
)
1367 /* if (cntntrecs >= maxntrecs) { only write every maxntrecs amount of data */
1370 gtperr
= GetTapePos(&curtapepos
);
1371 /* if an error occured previously don't
1375 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1377 gtperr
= MkTapeString(spclptr
, curtapepos
);
1383 #endif /* USE_QFA */
1385 while (eot_count
< 10 && size
< bufsize
) {
1388 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1391 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1393 printf("slave %d wrote %d\n", slave_number
, wrote
);
1395 if (wrote
< 0 && errno
!= ENOSPC
)
1397 if (wrote
== 0 || (wrote
< 0 && errno
== ENOSPC
))
1404 if (size
!= bufsize
)
1405 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1406 slave_number
, size
, bufsize
);
1410 * Handle ENOSPC as an EOT condition.
1412 if (wrote
< 0 && errno
== ENOSPC
) {
1418 returns
.clen
= returns
.unclen
= 0;
1421 * pass errno back to master for special handling
1424 returns
.unclen
= -errno
;
1427 * pass size of data and size of write back to master
1428 * (for EOT handling)
1430 (void) dump_atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1433 * Signal the next slave to go.
1435 (void) kill(nextslave
, SIGUSR2
);
1437 if (gTapeposfd
>= 0) {
1440 #endif /* USE_QFA */
1443 quit("error reading command pipe: %s\n", strerror(errno
));
1447 * Since a read from a pipe may not return all we asked for,
1448 * or a write may not write all we ask if we get a signal,
1449 * loop until the count is satisfied (or error).
1452 dump_atomic_read(int fd
, char *buf
, size_t count
)
1454 int got
, need
= count
;
1457 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1459 } while (got
== -1 && errno
== EINTR
);
1460 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1464 * Since a read from a pipe may not return all we asked for,
1465 * or a write may not write all we ask if we get a signal,
1466 * loop until the count is satisfied (or error).
1469 dump_atomic_write(int fd
, const char *buf
, size_t count
)
1471 int got
, need
= count
;
1474 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1476 } while (got
== -1 && errno
== EINTR
);
1477 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1488 buf.mt_op = MTSETDRVBUFFER;
1489 buf.mt_count = MT_ST_BOOLEANS | MT_ST_SCSI2LOGICAL;
1490 if (ioctl(tapefd, MTIOCTOP, &buf) == -1) {
1492 msg("[%ld] error: %d (setting logical)\n",
1493 (unsigned long)getpid(), err);
1500 #define LSEEK_GET_TAPEPOS 10
1501 #define LSEEK_GO2_TAPEPOS 11
1503 * read the current tape position
1506 GetTapePos(long long *pos
)
1512 *pos
= (long long) rmtseek((OFF_T
)0, (int)LSEEK_GET_TAPEPOS
);
1521 err
= (ioctl(tapefd
, MTIOCPOS
, &mtpos
) < 0);
1522 *pos
= (long long)mtpos
;
1525 *pos
= LSEEK(tapefd
, 0, SEEK_CUR
);
1531 msg("[%ld] error: %d (getting tapepos: %lld)\n", getpid(),
1539 MkTapeString(struct s_spcl
*spclptr
, long long curtapepos
)
1544 msg("inode %ld at tapepos %lld\n", spclptr
->c_inumber
, curtapepos
);
1547 snprintf(gTps
, sizeof(gTps
), "%ld\t%d\t%lld\n",
1548 (unsigned long)spclptr
->c_inumber
,
1551 gTps
[sizeof(gTps
) - 1] = '\0';
1552 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != (ssize_t
)strlen(gTps
)) {
1554 warn("error writing tapepos file. (error %d)\n", errno
);
1558 #endif /* USE_QFA */