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 1.94 2011/06/10 12:41:54 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 caught1
; /* have we caught the signal to proceed? */
182 static int ready1
; /* have we reached the lock point without having */
183 /* received the SIGUSR2 signal from the prev slave? */
184 static sigjmp_buf jmpbuf1
; /* where to jump to if we are ready when the */
185 /* SIGUSR1 arrives from the previous slave */
186 static int caught2
; /* have we caught the signal to proceed? */
187 static int ready2
; /* have we reached the lock point without having */
188 /* received the SIGUSR2 signal from the prev slave? */
189 static sigjmp_buf jmpbuf2
; /* where to jump to if we are ready when the */
190 /* SIGUSR2 arrives from the previous slave */
192 static int gtperr
= 0;
196 * Determine if we can use Linux' clone system call. If so, call it
197 * with the CLONE_IO flag so that all processes will share the same I/O
198 * context, allowing the I/O schedulers to make better scheduling decisions.
201 /* first, pull in the header files that define sys_clone and CLONE_IO */
208 /* If either is not present, fall back on the fork behaviour */
209 #if ! defined(SYS_clone) || ! defined (CLONE_IO)
210 #define fork_clone_io fork
211 #else /* SYS_clone */
212 /* CLONE_IO is available, determine which version of sys_clone to use */
213 #include <linux/version.h>
215 * Kernel 2.5.49 introduced two extra parameters to the clone system call.
216 * Neither is useful in our case, so this is easy to handle.
218 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,49)
220 * Parameters of the sys_clone syscall are
221 * clone_flags, child_stack, parent_tidptr, child_tidptr
222 * on all architectures except s390 and s390x
223 * s390* have child_stack, clone_flags, parent_tidptr, child_tidptr
225 #if defined(__s390__) || defined(__s390x__)
226 #define CLONE_ARGS 0, SIGCHLD|CLONE_IO, NULL, NULL
228 #define CLONE_ARGS SIGCHLD|CLONE_IO, 0, NULL, NULL
231 #define CLONE_ARGS SIGCHLD|CLONE_IO, 0
232 #endif /* LINUX_VERSION_CODE */
233 pid_t
fork_clone_io(void);
234 #endif /* SYS_clone */
235 #else /* __linux__ not defined */
236 #define fork_clone_io fork
237 #endif /* __linux__ */
242 int pgoff
= getpagesize() - 1;
246 writesize
= ntrec
* TP_BSIZE
;
247 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
249 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
250 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
251 * repositioning after stopping, i.e, streaming mode, where the gap is
252 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
254 if (!blocksperfiles
&& !unlimited
)
255 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
261 * Allocate tape buffer contiguous with the array of instruction
262 * packets, so flushtape() can write them together with one write().
263 * Align tape buffer on page boundary to speed up tape write().
265 for (i
= 0; i
<= SLAVES
; i
++) {
267 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
270 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
272 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
274 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
276 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
282 nextblock
= slp
->tblock
;
287 writerec(const void *dp
, int isspcl
)
290 slp
->req
[trecno
].dblk
= (ext2_loff_t
)0;
291 slp
->req
[trecno
].count
= 1;
292 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
293 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
295 /* Need to write it to the archive file */
296 if (! AfileActive
&& isspcl
&& (spcl
.c_type
== TS_END
))
298 if (AfileActive
&& Afile
>= 0 && !(spcl
.c_flags
& DR_EXTATTRIBUTES
)) {
299 /* When we dump an inode which is not a directory,
300 * it means we ended the archive contents */
301 if (isspcl
&& (spcl
.c_type
== TS_INODE
) &&
302 ((spcl
.c_dinode
.di_mode
& S_IFMT
) != IFDIR
))
306 tmp
= *(union u_spcl
*)dp
;
307 /* Write the record, _uncompressed_ */
309 tmp
.s_spcl
.c_flags
&= ~DR_COMPRESSED
;
312 if (write(Afile
, &tmp
, TP_BSIZE
) != TP_BSIZE
)
313 msg("error writing archive file: %s\n",
320 lastspclrec
= spcl
.c_tapea
;
328 dumpblock(blk_t blkno
, int size
)
333 dblkno
= fsbtodb(sblock
, blkno
);
334 tpblks
= size
>> tp_bshift
;
335 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
336 slp
->req
[trecno
].dblk
= dblkno
;
337 slp
->req
[trecno
].count
= avail
;
339 spcl
.c_tapea
+= avail
;
342 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
354 msg("write error on %s: %s\n", tape
, strerror(errnum
));
355 quit("Cannot recover\n");
358 msg("write error %d blocks into volume %d: %s\n",
359 blocksthisvol
, tapeno
, strerror(errnum
));
360 broadcast("DUMP WRITE ERROR!\n");
361 if (query("Do you want to rewrite this volume?")) {
362 msg("Closing this volume. Prepare to restart with new media;\n");
363 msg("this dump volume will be rewritten.\n");
369 if (query("Do you want to start the next tape?"))
375 sigpipe(UNUSED(int signo
))
378 quit("Broken pipe\n");
383 * Update xferrate stats
392 ttaken
= tnow
- tstart_volume
;
393 blocks
= spcl
.c_tapea
- tapea_volume
;
394 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
396 msg("Volume %d %ld blocks (%.2fMB)\n", tapeno
,
397 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
399 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
400 long txfrate
= volkb
/ ttaken
;
401 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
402 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
403 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
407 double rate
= .0005 + (double) blocks
/ (double) volkb
;
408 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
410 tapeno
, blocks
, volkb
, rate
);
417 mktimeest(time_t tnow
)
419 static char msgbuf
[128];
424 if (blockswritten
< 500)
426 if (blockswritten
> tapesize
)
427 tapesize
= blockswritten
;
428 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
429 / blockswritten
* tapesize
;
430 if (tnow
> tstart_volume
)
431 (void)snprintf(msgbuf
, sizeof(msgbuf
),
432 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
433 (blockswritten
* 100.0) / tapesize
,
434 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
435 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
437 (void)snprintf(msgbuf
, sizeof(msgbuf
),
438 "%3.2f%% done, finished in %d:%02d\n",
439 (blockswritten
* 100.0) / tapesize
,
440 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
448 * information message upon receipt of SIGINFO
451 statussig(int notused
)
453 int save_errno
= errno
;
456 buf
= mktimeest(time(NULL
));
458 write(STDERR_FILENO
, buf
, strlen(buf
));
468 struct slave_results returned
;
470 int siz
= (char *)nextblock
- (char *)slp
->req
;
472 /* make sure returned has sane values in case we don't read
473 * them from the slave in this pass */
474 returned
.unclen
= returned
.clen
= writesize
;
476 slp
->req
[trecno
].count
= 0; /* Sentinel */
478 if (dump_atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
479 quit("error writing command pipe: %s\n", strerror(errno
));
480 slp
->sent
= 1; /* we sent a request, read the response later */
482 lastfirstrec
= slp
->firstrec
;
484 if (++slp
>= &slaves
[SLAVES
])
487 /* Read results back from next slave */
489 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
490 != sizeof returned
) {
491 perror(" DUMP: error reading command pipe in master");
494 got
= returned
.unclen
;
495 bytes_written
+= returned
.clen
;
496 if (returned
.unclen
== returned
.clen
)
500 /* Check for errors or end of tape */
502 /* Check for errors */
506 msg("End of tape detected\n");
509 * Drain the results, don't care what the values were.
510 * If we read them here then trewind won't...
512 for (i
= 0; i
< SLAVES
; i
++) {
513 if (slaves
[i
].sent
) {
514 if (dump_atomic_read( slaves
[i
].fd
,
515 (char *)&returned
, sizeof returned
)
516 != sizeof returned
) {
517 perror(" DUMP: error reading command pipe in master");
531 if (spcl
.c_type
== TS_CLRI
|| spcl
.c_type
== TS_BITS
)
534 if (spcl
.c_type
!= TS_END
) {
535 for (i
= 0; i
< spcl
.c_count
; i
++)
536 if (spcl
.c_addr
[i
] != 0)
540 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
541 slp
->tapea
= spcl
.c_tapea
;
542 slp
->firstrec
= lastfirstrec
+ ntrec
;
544 nextblock
= slp
->tblock
;
546 asize
+= tenths
+ returned
.clen
/ density
;
547 blockswritten
+= ntrec
;
548 blocksthisvol
+= ntrec
;
549 if (!pipeout
&& !unlimited
) {
550 if (blocksperfiles
&& blocksperfiles
[blocksperfiles_current
]) {
551 if ( compressed
? (bytes_written
- tapea_bytes
+ SLAVES
* (writesize
+ sizeof(struct tapebuf
))) >= (((long long)blocksperfiles
[blocksperfiles_current
]) * 1024)
552 : blocksthisvol
>= blocksperfiles
[blocksperfiles_current
] ) {
557 else if (asize
> tsize
) {
570 struct slave_results returned
;
572 for (f
= 0; f
< SLAVES
; f
++) {
574 * Drain the results, but unlike EOT we DO (or should) care
575 * what the return values were, since if we detect EOT after
576 * we think we've written the last blocks to the tape anyway,
577 * we have to replay those blocks with rollforward.
579 * fixme: punt for now.
581 if (slaves
[f
].sent
) {
582 if (dump_atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
583 != sizeof returned
) {
584 perror(" DUMP: error reading command pipe in master");
587 got
= returned
.unclen
;
588 bytes_written
+= returned
.clen
;
589 if (returned
.unclen
== returned
.clen
)
597 msg("EOT detected in last 2 tape records!\n");
598 msg("Use a longer tape, decrease the size estimate\n");
599 quit("or use no size estimate at all.\n");
602 (void) close(slaves
[f
].fd
);
604 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
609 msg("Closing %s\n", tape
);
614 while (rmtopen(tape
, O_RDONLY
) < 0)
621 (void) close(tapefd
);
623 while ((f
= OPEN(tape
, O_RDONLY
)) < 0)
639 msg("Launching %s\n", eot_script
);
640 eot_code
= system_command(eot_script
, tape
, tapeno
);
642 if (eot_code
!= 0 && eot_code
!= 1) {
643 msg("Dump aborted by the end of tape script\n");
648 if (nexttape
|| Mflag
)
651 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
652 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
654 while (!query("Is the new volume mounted and ready to go?"))
655 if (query("Do you want to abort?")) {
664 struct req
*p
, *q
= NULL
, *prev
;
666 int i
, size
, savedtapea
, got
;
667 union u_spcl
*ntb
, *otb
;
668 struct slave_results returned
;
673 tslp
= &slaves
[SLAVES
];
674 ntb
= (union u_spcl
*)tslp
->tblock
[1];
676 /* make sure returned has sane values in case we don't read
677 * them from the slave in this pass */
678 returned
.unclen
= returned
.clen
= writesize
;
681 * Each of the N slaves should have requests that need to
682 * be replayed on the next tape. Use the extra slave buffers
683 * (slaves[SLAVES]) to construct request lists to be sent to
684 * each slave in turn.
686 for (i
= 0; i
< SLAVES
; i
++) {
688 otb
= (union u_spcl
*)slp
->tblock
;
691 * For each request in the current slave, copy it to tslp.
695 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
698 *ntb
++ = *otb
++; /* copy the datablock also */
703 quit("rollforward: protocol botch");
715 nextblock
= tslp
->tblock
;
716 savedtapea
= spcl
.c_tapea
;
717 spcl
.c_tapea
= slp
->tapea
;
719 spcl
.c_tapea
= savedtapea
;
720 lastspclrec
= savedtapea
- 1;
722 size
= (char *)ntb
- (char *)q
;
723 if (dump_atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
724 perror(" DUMP: error writing command pipe");
729 lastfirstrec
= slp
->firstrec
;
731 if (++slp
>= &slaves
[SLAVES
])
736 if (prev
->dblk
!= 0) {
738 * If the last one was a disk block, make the
739 * first of this one be the last bit of that disk
742 q
->dblk
= prev
->dblk
+
743 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
744 ntb
= (union u_spcl
*)tslp
->tblock
;
747 * It wasn't a disk block. Copy the data to its
748 * new location in the buffer.
751 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
752 ntb
= (union u_spcl
*)tslp
->tblock
[1];
756 nextblock
= slp
->tblock
;
759 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
760 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
767 * Clear the first slaves' response. One hopes that it
768 * worked ok, otherwise the tape is much too short!
771 if (dump_atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
772 != sizeof returned
) {
773 perror(" DUMP: error reading command pipe in master");
776 got
= returned
.unclen
;
777 bytes_written
+= returned
.clen
;
778 if (returned
.clen
== returned
.unclen
)
786 quit("EOT detected at start of the tape!\n");
792 if (spcl
.c_type
!= TS_END
) {
793 for (i
= 0; i
< spcl
.c_count
; i
++)
794 if (spcl
.c_addr
[i
] != 0)
798 slp
->firstrec
= lastfirstrec
+ ntrec
;
799 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
801 asize
+= tenths
+ returned
.clen
/ density
;
802 blockswritten
+= ntrec
;
803 blocksthisvol
+= ntrec
;
808 #if defined(SYS_clone) && defined(CLONE_IO)
812 return syscall(SYS_clone
, CLONE_ARGS
);
818 * We implement taking and restoring checkpoints on the tape level.
819 * When each tape is opened, a new process is created by forking; this
820 * saves all of the necessary context in the parent. The child
821 * continues the dump; the parent waits around, saving the context.
822 * If the child returns X_REWRITE, then it had problems writing that tape;
823 * this causes the parent to fork again, duplicating the context, and
824 * everything continues as if nothing had happened.
827 startnewtape(int top
)
838 sigaddset(&sigs
, SIGINT
);
839 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
840 #else /* __linux__ */
842 void (*interrupt_save
)();
844 sig_t interrupt_save
;
846 interrupt_save
= signal(SIGINT
, SIG_IGN
);
847 #endif /* __linux__ */
849 parentpid
= getpid();
850 tapea_volume
= spcl
.c_tapea
;
851 tapea_bytes
= bytes_written
;
852 tstart_volume
= time(NULL
);
856 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
858 (void)signal(SIGINT
, interrupt_save
);
861 * All signals are inherited...
863 childpid
= fork_clone_io();
865 msg("Context save fork fails in parent %d\n", parentpid
);
871 * save the context by waiting
872 * until the child doing all of the work returns.
873 * don't catch the interrupt
876 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
878 signal(SIGINT
, SIG_IGN
);
881 msg("Tape: %d; parent process: %d child process %d\n",
882 tapeno
+1, parentpid
, childpid
);
884 while ((waitpid
= wait(&status
)) != childpid
)
885 if (waitpid
!= rshpid
)
886 msg("Parent %d waiting for child %d has another child %d return\n",
887 parentpid
, childpid
, waitpid
);
889 msg("Child %d returns LOB status %o\n",
890 childpid
, status
&0xFF);
892 status
= (status
>> 8) & 0xFF;
896 msg("Child %d finishes X_FINOK\n", childpid
);
899 msg("Child %d finishes X_ABORT\n", childpid
);
902 msg("Child %d finishes X_REWRITE\n", childpid
);
905 msg("Child %d finishes unknown %d\n",
916 goto restore_check_point
;
918 msg("Bad return code from dump: %d\n", status
);
922 } else { /* we are the child; just continue */
924 sleep(4); /* allow time for parent's message to get out */
925 msg("Child on Tape %d has parent %d, my pid = %d\n",
926 tapeno
+1, parentpid
, getpid());
929 * If we have a name like "/dev/rmt0,/dev/rmt1",
930 * use the name before the comma first, and save
931 * the remaining names for subsequent volumes.
933 tapeno
++; /* current tape sequence */
935 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
936 tape
[MAXPATHLEN
- 1] = '\0';
937 msg("Dumping volume %d on %s\n", tapeno
, tape
);
939 else if (nexttape
|| strchr(tapeprefix
, ',')) {
940 if (nexttape
&& *nexttape
)
941 tapeprefix
= nexttape
;
942 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
947 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
948 tape
[MAXPATHLEN
- 1] = '\0';
949 msg("Dumping volume %d on %s\n", tapeno
, tape
);
951 if (blocksperfiles
&& blocksperfiles_current
< *blocksperfiles
)
952 blocksperfiles_current
++;
954 while ((tapefd
= (host
? rmtopen(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
) : pipeout
?
956 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
958 while ((tapefd
= (pipeout
? fileno(stdout
) :
959 OPEN(tape
, O_WRONLY
|O_CREAT
|O_TRUNC
, 0666))) < 0)
962 msg("Cannot open output \"%s\": %s\n", tape
,
964 if (!query("Do you want to retry the open?"))
971 struct mtget mt_stat
;
972 magtapeout
= ioctl(tapefd
, MTIOCGET
, (char *)&mt_stat
) == 0;
974 msg("Output is to %s\n",
975 magtapeout ? "tape" : "file/pipe");
979 enslave(); /* Share open tape file descriptor with slaves */
984 newtape
++; /* new tape signal */
985 spcl
.c_count
= slp
->count
;
987 * measure firstrec in TP_BSIZE units since restore doesn't
988 * know the correct ntrec value...
990 spcl
.c_firstrec
= slp
->firstrec
;
992 spcl
.c_type
= TS_TAPE
;
993 spcl
.c_flags
|= DR_NEWHEADER
;
994 spcl
.c_ntrec
= ntrec
;
996 spcl
.c_flags
|= DR_COMPRESSED
;
997 writeheader((dump_ino_t
)slp
->inode
);
998 spcl
.c_flags
&=~ DR_NEWHEADER
;
999 msg("Volume %d started with block %ld at: %s", tapeno
,
1000 spcl
.c_tapea
, ctime(&tstart_volume
));
1002 msg("Volume %d begins with blocks from inode %d\n",
1003 tapeno
, slp
->inode
);
1004 if (tapeno
< (int)TP_NINOS
)
1005 volinfo
[tapeno
] = slp
->inode
;
1010 dumpabort(UNUSED(int signo
))
1013 if (master
!= 0 && master
!= getpid())
1014 /* Signals master to call dumpabort */
1015 (void) kill(master
, SIGTERM
);
1018 msg("The ENTIRE dump is aborted.\n");
1031 msg("pid = %d exits with status %d\n", getpid(), status
);
1037 * proceed - handler for SIGUSR1, used to synchronize IO between the slaves.
1040 proceed1(UNUSED(int signo
))
1043 siglongjmp(jmpbuf1
, 1);
1048 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
1051 proceed2(UNUSED(int signo
))
1054 siglongjmp(jmpbuf2
, 1);
1062 #ifdef LINUX_FORK_BUG
1070 { struct sigaction sa
;
1071 memset(&sa
, 0, sizeof sa
);
1072 sigemptyset(&sa
.sa_mask
);
1073 sa
.sa_handler
= dumpabort
;
1074 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
1075 sa
.sa_handler
= sigpipe
;
1076 sigaction(SIGPIPE
, &sa
, NULL
);
1077 sa
.sa_handler
= proceed1
;
1078 sa
.sa_flags
= SA_RESTART
;
1079 sigaction(SIGUSR1
, &sa
, NULL
); /* Slave sends SIGUSR1 to next slave */
1080 sa
.sa_handler
= proceed2
;
1081 sa
.sa_flags
= SA_RESTART
;
1082 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
1085 for (i
= 0; i
< SLAVES
; i
++) {
1086 if (i
== slp
- &slaves
[0]) {
1094 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
1095 (slaves
[i
].pid
= fork_clone_io()) < 0)
1096 quit("too many slaves, %d (recompile smaller): %s\n",
1097 i
, strerror(errno
));
1099 slaves
[i
].fd
= cmd
[1];
1101 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
1103 for (j
= 0; j
<= i
; j
++)
1104 (void) close(slaves
[j
].fd
);
1106 sigaddset(&sigs
, SIGINT
); /* Master handles this */
1107 #if defined(SIGINFO)
1108 sigaddset(&sigs
, SIGINFO
);
1110 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
1112 #ifdef LINUX_FORK_BUG
1113 if (dump_atomic_write( cmd
[0], (char *) &i
, sizeof i
)
1115 quit("master/slave protocol botched 3\n");
1121 (slaves
[i
].pid
== slp
->pid
));
1128 #ifdef LINUX_FORK_BUG
1130 * Wait for all slaves to _actually_ start to circumvent a bug in
1131 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1132 * returned from fork() causes a SEGV in the child process
1134 for (i
= 0; i
< SLAVES
; i
++)
1135 if (dump_atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1136 quit("master/slave protocol botched 4\n");
1139 for (i
= 0; i
< SLAVES
; i
++)
1140 (void) dump_atomic_write( slaves
[i
].fd
,
1141 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1142 sizeof slaves
[0].pid
);
1152 for (i
= 0; i
< SLAVES
; i
++)
1153 if (slaves
[i
].pid
> 0) {
1154 (void) kill(slaves
[i
].pid
, SIGKILL
);
1160 * Synchronization - each process waits for a SIGUSR2 from the
1161 * previous process before writing to the tape, and sends SIGUSR2
1162 * to the next process when the tape write completes. On tape errors
1163 * a SIGUSR1 is sent to the master which then terminates all of the
1164 * slaves. Each process sends SIGUSR1 to the next to signal that it
1165 * is time to start reading from the disk, after it finishes reading
1166 * and moves to the compression phase.
1177 volatile int wrote
= 0, size
, eot_count
, bufsize
;
1178 char * volatile buffer
;
1179 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1180 struct tapebuf
* volatile comp_buf
= NULL
;
1182 volatile int do_compress
= !first
;
1183 unsigned long worklen
;
1185 lzo_align_t __LZO_MMODEL
*LZO_WorkMem
;
1187 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1188 struct slave_results returns
;
1193 long long curtapepos
;
1194 union u_spcl
*uspclptr
;
1195 struct s_spcl
*spclptr
;
1196 /* long maxntrecs = 300000000 / (ntrec * 1024); last tested: 50 000 000 */
1197 long maxntrecs
= 50000; /* every 50MB */
1198 long cntntrecs
= maxntrecs
;
1199 #endif /* USE_QFA */
1203 sigaddset(&set
, SIGUSR1
);
1204 sigaddset(&set
, SIGUSR2
);
1205 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1209 * Need our own seek pointer.
1211 (void) close(diskfd
);
1212 if ((diskfd
= OPEN(disk
, O_RDONLY
)) < 0)
1213 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1216 (void)ioctl(diskfd
, BLKFLSBUF
, 0);
1219 retval
= dump_fs_open(disk
, &fs
);
1221 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1222 #endif /* __linux__ */
1225 * Need the pid of the next slave in the loop...
1227 if ((nread
= dump_atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1228 != sizeof nextslave
) {
1229 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1232 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1233 /* if we're doing a compressed dump, allocate the compress buffer */
1235 int bsiz
= sizeof(struct tapebuf
) + writesize
;
1236 /* Add extra space to deal with compression enlarging the buffer */
1237 if (TP_BSIZE
> writesize
/16 + 67)
1240 bsiz
+= writesize
/16 + 67;
1241 comp_buf
= malloc(bsiz
);
1242 if (comp_buf
== NULL
)
1243 quit("couldn't allocate a compress buffer.\n");
1244 if (zipflag
== COMPRESS_ZLIB
)
1245 comp_buf
->flags
= COMPRESS_ZLIB
;
1246 else if (zipflag
== COMPRESS_BZLIB
)
1247 comp_buf
->flags
= COMPRESS_BZLIB
;
1248 else if (zipflag
== COMPRESS_LZO
) {
1249 comp_buf
->flags
= COMPRESS_LZO
;
1250 if (lzo_init() != LZO_E_OK
) quit("lzo_init failed\n");
1252 quit("internal error - unknown compression method: %d\n", zipflag
);
1255 LZO_WorkMem
= malloc(LZO1X_1_MEM_COMPRESS
);
1257 quit("couldn't allocate a compress buffer.\n");
1259 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1262 * Get list of blocks to dump, read the blocks into tape buffer
1264 while ((nread
= dump_atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1265 struct req
*p
= slp
->req
;
1267 /* wait for previous slave to finish reading */
1268 if (sigsetjmp(jmpbuf1
, 1) == 0) {
1276 for (trecno
= 0; trecno
< ntrec
;
1277 trecno
+= p
->count
, p
+= p
->count
) {
1278 if (p
->dblk
) { /* read a disk block */
1279 bread(p
->dblk
, slp
->tblock
[trecno
],
1280 p
->count
* TP_BSIZE
);
1281 } else { /* read record from pipe */
1282 if (p
->count
!= 1 || dump_atomic_read( cmd
,
1283 (char *)slp
->tblock
[trecno
],
1284 TP_BSIZE
) != TP_BSIZE
)
1285 quit("master/slave protocol botched.\n");
1288 /* signal next slave to start reading */
1289 (void) kill(nextslave
, SIGUSR1
);
1291 /* Try to write the data... */
1295 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1296 bufsize
= writesize
; /* length to write */
1297 returns
.clen
= returns
.unclen
= bufsize
;
1299 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB) || defined(HAVE_LZO)
1301 * When writing a compressed dump, each block except
1302 * the first one on each tape is written
1303 * from struct tapebuf with an 4 byte prefix
1304 * followed by the data. This can be less than
1305 * writesize. Restore, on a short read, can compare the
1306 * length read to the compressed length in the header
1307 * to verify that the read was good. Blocks which don't
1308 * compress well are written uncompressed.
1309 * The first block written by each slave is not compressed
1310 * and does not have a prefix.
1313 if (compressed
&& do_compress
) {
1314 comp_buf
->length
= bufsize
;
1315 worklen
= TP_BSIZE
+ writesize
;
1318 if (zipflag
== COMPRESS_ZLIB
) {
1319 compresult
= compress2(comp_buf
->buf
,
1321 (char *)slp
->tblock
[0],
1324 if (compresult
== Z_OK
)
1329 #endif /* HAVE_ZLIB */
1331 if (zipflag
== COMPRESS_BZLIB
) {
1332 unsigned int worklen2
= worklen
;
1333 compresult
= BZ2_bzBuffToBuffCompress(
1336 (char *)slp
->tblock
[0],
1341 if (compresult
== BZ_OK
)
1347 #endif /* HAVE_BZLIB */
1349 if (zipflag
== COMPRESS_LZO
) {
1350 lzo_uint worklen2
= worklen
;
1351 compresult
= lzo1x_1_compress((char *)slp
->tblock
[0],writesize
,
1356 if (compresult
== LZO_E_OK
)
1361 #endif /* HAVE_LZO */
1362 if (compresult
&& worklen
<= ((unsigned long)writesize
- 16)) {
1363 /* write the compressed buffer */
1364 comp_buf
->length
= worklen
;
1365 comp_buf
->compressed
= 1;
1366 buffer
= (char *) comp_buf
;
1367 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1370 /* write the data uncompressed */
1371 comp_buf
->length
= writesize
;
1372 comp_buf
->compressed
= 0;
1373 buffer
= (char *) comp_buf
;
1374 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1375 returns
.unclen
= returns
.clen
;
1376 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1379 /* compress the remaining blocks if we're compressing */
1380 do_compress
= compressed
;
1381 #endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
1383 if (sigsetjmp(jmpbuf2
, 1) == 0) {
1392 if (gTapeposfd
>= 0) {
1396 for (i
= 0; (i
< ntrec
) && !foundone
; ++i
) {
1397 uspclptr
= (union u_spcl
*)&slp
->tblock
[i
];
1398 spclptr
= &uspclptr
->s_spcl
;
1399 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1400 (spclptr
->c_type
== TS_INODE
) &&
1401 (spclptr
->c_date
== gThisDumpDate
) &&
1402 !(spclptr
->c_dinode
.di_mode
& S_IFDIR
) &&
1403 !(spclptr
->c_flags
& DR_EXTATTRIBUTES
)
1406 /* if (cntntrecs >= maxntrecs) { only write every maxntrecs amount of data */
1409 gtperr
= GetTapePos(&curtapepos
);
1410 /* if an error occured previously don't
1414 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1416 gtperr
= MkTapeString(spclptr
, curtapepos
);
1422 #endif /* USE_QFA */
1424 while (eot_count
< 10 && size
< bufsize
) {
1427 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1430 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1432 printf("slave %d wrote %d\n", slave_number
, wrote
);
1434 if (wrote
< 0 && errno
!= ENOSPC
)
1436 if (wrote
== 0 || (wrote
< 0 && errno
== ENOSPC
))
1443 if (size
!= bufsize
)
1444 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1445 slave_number
, size
, bufsize
);
1449 * Handle ENOSPC as an EOT condition.
1451 if (wrote
< 0 && errno
== ENOSPC
) {
1457 returns
.clen
= returns
.unclen
= 0;
1460 * pass errno back to master for special handling
1463 returns
.unclen
= -errno
;
1466 * pass size of data and size of write back to master
1467 * (for EOT handling)
1469 (void) dump_atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1472 * Signal the next slave to go.
1474 (void) kill(nextslave
, SIGUSR2
);
1476 if (gTapeposfd
>= 0) {
1479 #endif /* USE_QFA */
1482 quit("error reading command pipe: %s\n", strerror(errno
));
1486 * Since a read from a pipe may not return all we asked for,
1487 * or a write may not write all we ask if we get a signal,
1488 * loop until the count is satisfied (or error).
1491 dump_atomic_read(int fd
, char *buf
, size_t count
)
1493 int got
, need
= count
;
1496 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1498 } while (got
== -1 && errno
== EINTR
);
1499 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1503 * Since a read from a pipe may not return all we asked for,
1504 * or a write may not write all we ask if we get a signal,
1505 * loop until the count is satisfied (or error).
1508 dump_atomic_write(int fd
, const char *buf
, size_t count
)
1510 int got
, need
= count
;
1513 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1515 } while (got
== -1 && errno
== EINTR
);
1516 return (got
< 0 ? got
: (ssize_t
)count
- need
);
1527 buf.mt_op = MTSETDRVBUFFER;
1528 buf.mt_count = MT_ST_BOOLEANS | MT_ST_SCSI2LOGICAL;
1529 if (ioctl(tapefd, MTIOCTOP, &buf) == -1) {
1531 msg("[%ld] error: %d (setting logical)\n",
1532 (unsigned long)getpid(), err);
1539 #define LSEEK_GET_TAPEPOS 10
1540 #define LSEEK_GO2_TAPEPOS 11
1542 * read the current tape position
1545 GetTapePos(long long *pos
)
1551 *pos
= (long long) rmtseek((OFF_T
)0, (int)LSEEK_GET_TAPEPOS
);
1560 err
= (ioctl(tapefd
, MTIOCPOS
, &mtpos
) < 0);
1561 *pos
= (long long)mtpos
;
1564 *pos
= LSEEK(tapefd
, 0, SEEK_CUR
);
1570 msg("[%ld] error: %d (getting tapepos: %lld)\n", getpid(),
1578 MkTapeString(struct s_spcl
*spclptr
, long long curtapepos
)
1583 msg("inode %ld at tapepos %lld\n", spclptr
->c_inumber
, curtapepos
);
1586 snprintf(gTps
, sizeof(gTps
), "%ld\t%d\t%lld\n",
1587 (unsigned long)spclptr
->c_inumber
,
1590 gTps
[sizeof(gTps
) - 1] = '\0';
1591 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != (ssize_t
)strlen(gTps
)) {
1593 warn("error writing tapepos file. (error %d)\n", errno
);
1597 #endif /* USE_QFA */