2 * Ported to Linux's Second Extended File System as part of the
3 * dump and restore backup suit
4 * Remy Card <card@Linux.EU.Org>, 1994-1997
5 * Stelian Pop <stelian@popies.net>, 1999-2000
6 * Stelian Pop <stelian@popies.net> - AlcĂ´ve <www.alcove.com>, 2000-2002
10 * Copyright (c) 1980, 1991, 1993
11 * The Regents of the University of California. All rights reserved.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 static const char rcsid[] =
44 "$Id: tape.c,v 1.63 2002/02/21 09:25:31 stelian Exp $";
48 #include <compatlfs.h>
54 #include <compaterr.h>
65 #include <sys/types.h>
69 #include <sys/param.h>
70 #include <sys/socket.h>
75 #undef atomic_read /* this get wrongly defined in kernel */
76 /* headers and we don't want it */
77 #ifdef HAVE_EXT2FS_EXT2_FS_H
78 #include <ext2fs/ext2_fs.h>
80 #include <linux/ext2_fs.h>
82 #include <ext2fs/ext2fs.h>
84 #include <bsdcompat.h>
86 #include <sys/vnode.h>
89 #include <ufs/inode.h>
91 #include <ufs/ufs/dinode.h>
92 #include <ufs/ffs/fs.h>
93 #endif /* __linux__ */
95 #include <protocols/dumprestore.h>
99 #endif /* HAVE_ZLIB */
103 #endif /* HAVE_BZLIB */
107 int writesize; /* size of malloc()ed buffer for tape */
108 long lastspclrec = -1; /* tape block number of last written header */
109 int trecno = 0; /* next record to write in current block */
110 extern long blocksperfile; /* number of blocks per output file */
111 long blocksthisvol; /* number of blocks on current output file */
112 extern int ntrec; /* blocking factor on tape */
113 extern int cartridge;
117 long long tapea_bytes = 0; /* bytes_written at start of current volume */
118 static int magtapeout; /* output is really a tape */
120 static ssize_t atomic_read __P((int, void *, size_t));
121 static ssize_t atomic_write __P((int, const void *, size_t));
122 static void doslave __P((int, int, int));
123 static void enslave __P((void));
124 static void flushtape __P((void));
125 static void killall __P((void));
126 static void rollforward __P((void));
128 static int GetTapePos __P((long long *));
129 static void MkTapeString __P((struct s_spcl *, long long));
133 * Concurrent dump mods (Caltech) - disk block reading and tape writing
134 * are exported to several slave processes. While one slave writes the
135 * tape, the others read disk blocks; they pass control of the tape in
136 * a ring via signals. The parent process traverses the filesystem and
137 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
138 * The following structure defines the instruction packets sent to slaves.
146 struct slave_results {
147 ssize_t unclen; /* uncompressed length */
148 ssize_t clen; /* compressed length */
151 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
153 int tapea; /* header number at start of this chunk */
154 int count; /* count to next header (used for TS_TAPE */
156 int inode; /* inode that we are currently dealing with */
157 int fd; /* FD for this slave */
158 int pid; /* PID for this slave */
159 int sent; /* 1 == we've sent this slave requests */
160 int firstrec; /* record number of this block */
161 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
162 struct req *req; /* buffer for requests */
166 char (*nextblock)[TP_BSIZE];
168 static time_t tstart_volume; /* time of volume start */
169 static int tapea_volume; /* value of spcl.c_tapea at volume start */
171 int master; /* pid of master, for sending error signals */
172 int tenths; /* length of tape overhead per block written */
173 static int caught; /* have we caught the signal to proceed? */
174 static int ready; /* have we reached the lock point without having */
175 /* received the SIGUSR2 signal from the prev slave? */
176 static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */
177 /* SIGUSR2 arrives from the previous slave */
179 static int gtperr = 0;
185 int pgoff = getpagesize() - 1;
189 writesize = ntrec * TP_BSIZE;
190 reqsiz = (ntrec + 1) * sizeof(struct req);
192 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
193 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
194 * repositioning after stopping, i.e, streaming mode, where the gap is
195 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
197 if (blocksperfile == 0 && !unlimited)
198 tenths = (cartridge ? 16 : density == 625 ? 5 : 8);
204 * Allocate tape buffer contiguous with the array of instruction
205 * packets, so flushtape() can write them together with one write().
206 * Align tape buffer on page boundary to speed up tape write().
208 for (i = 0; i <= SLAVES; i++) {
210 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
213 slaves[i].tblock = (char (*)[TP_BSIZE])
215 (((long)&buf[reqsiz] + pgoff) &~ pgoff);
217 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
219 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
225 nextblock = slp->tblock;
230 writerec(const void *dp, int isspcl)
233 slp->req[trecno].dblk = (daddr_t)0;
234 slp->req[trecno].count = 1;
235 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
236 *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp;
238 /* Need to write it to the archive file */
239 if (Afile < 0 && isspcl && (spcl.c_type == TS_END))
242 /* When we dump an inode which is not a directory,
243 * it means we ended the archive contents */
244 if (isspcl && (spcl.c_type == TS_INODE) &&
245 ((spcl.c_dinode.di_mode & S_IFMT) != IFDIR))
249 tmp = *(union u_spcl *)dp;
250 /* Write the record, _uncompressed_ */
252 tmp.s_spcl.c_flags &= ~DR_COMPRESSED;
255 if (write(Afile, &tmp, TP_BSIZE) != TP_BSIZE)
256 msg("error writing archive file: %s\n",
263 lastspclrec = spcl.c_tapea;
271 dumpblock(daddr_t blkno, int size)
273 int avail, tpblks, dblkno;
275 dblkno = fsbtodb(sblock, blkno);
276 tpblks = size >> tp_bshift;
277 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
278 slp->req[trecno].dblk = dblkno;
279 slp->req[trecno].count = avail;
281 spcl.c_tapea += avail;
284 dblkno += avail << (tp_bshift - dev_bshift);
296 msg("write error on %s: %s\n", tape, strerror(errnum));
297 quit("Cannot recover\n");
300 msg("write error %d blocks into volume %d: %s\n",
301 blocksthisvol, tapeno, strerror(errnum));
302 broadcast("DUMP WRITE ERROR!\n");
303 if (query("Do you want to rewrite this volume?")) {
304 msg("Closing this volume. Prepare to restart with new media;\n");
305 msg("this dump volume will be rewritten.\n");
311 if (query("Do you want to start the next tape?"))
320 quit("Broken pipe\n");
325 * Update xferrate stats
334 ttaken = tnow - tstart_volume;
335 blocks = spcl.c_tapea - tapea_volume;
336 msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
338 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno,
339 blocks, ((double)blocks * TP_BSIZE / 1048576));
341 long volkb = (bytes_written - tapea_bytes) / 1024;
342 long txfrate = volkb / ttaken;
343 msg("Volume %d took %d:%02d:%02d\n", tapeno,
344 ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60);
345 msg("Volume %d transfer rate: %ld kB/s\n", tapeno,
349 double rate = .0005 + (double) blocks / (double) volkb;
350 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
352 tapeno, blocks, volkb, rate);
359 mktimeest(time_t tnow)
361 static char msgbuf[128];
366 if (blockswritten < 500)
368 if (blockswritten > tapesize)
369 tapesize = blockswritten;
370 deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing))
371 / blockswritten * tapesize;
372 if (tnow > tstart_volume)
373 (void)snprintf(msgbuf, sizeof(msgbuf),
374 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
375 (blockswritten * 100.0) / tapesize,
376 (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume),
377 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
379 (void)snprintf(msgbuf, sizeof(msgbuf),
380 "%3.2f%% done, finished in %d:%02d\n",
381 (blockswritten * 100.0) / tapesize,
382 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
390 * information message upon receipt of SIGINFO
393 statussig(int notused)
395 int save_errno = errno;
398 buf = mktimeest(time(NULL));
400 write(STDERR_FILENO, buf, strlen(buf));
410 struct slave_results returned;
412 int siz = (char *)nextblock - (char *)slp->req;
414 slp->req[trecno].count = 0; /* Sentinel */
416 if (atomic_write( slp->fd, (char *)slp->req, siz) != siz)
417 quit("error writing command pipe: %s\n", strerror(errno));
418 slp->sent = 1; /* we sent a request, read the response later */
420 lastfirstrec = slp->firstrec;
422 if (++slp >= &slaves[SLAVES])
425 /* Read results back from next slave */
427 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
428 != sizeof returned) {
429 perror(" DUMP: error reading command pipe in master");
432 got = returned.unclen;
433 bytes_written += returned.clen;
434 if (returned.unclen == returned.clen)
438 /* Check for errors or end of tape */
440 /* Check for errors */
444 msg("End of tape detected\n");
447 * Drain the results, don't care what the values were.
448 * If we read them here then trewind won't...
450 for (i = 0; i < SLAVES; i++) {
451 if (slaves[i].sent) {
452 if (atomic_read( slaves[i].fd,
453 (char *)&returned, sizeof returned)
454 != sizeof returned) {
455 perror(" DUMP: error reading command pipe in master");
469 if (spcl.c_type != TS_END) {
470 for (i = 0; i < spcl.c_count; i++)
471 if (spcl.c_addr[i] != 0)
474 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
475 slp->tapea = spcl.c_tapea;
476 slp->firstrec = lastfirstrec + ntrec;
478 nextblock = slp->tblock;
480 asize += tenths + returned.clen / density;
481 blockswritten += ntrec;
482 blocksthisvol += ntrec;
483 if (!pipeout && !unlimited) {
485 if ( compressed ? (bytes_written - tapea_bytes + SLAVES * (writesize + sizeof(struct tapebuf))) >= blocksperfile * 1024
486 : blocksthisvol >= blocksperfile ) {
491 else if (asize > tsize) {
504 struct slave_results returned;
506 for (f = 0; f < SLAVES; f++) {
508 * Drain the results, but unlike EOT we DO (or should) care
509 * what the return values were, since if we detect EOT after
510 * we think we've written the last blocks to the tape anyway,
511 * we have to replay those blocks with rollforward.
513 * fixme: punt for now.
515 if (slaves[f].sent) {
516 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
517 != sizeof returned) {
518 perror(" DUMP: error reading command pipe in master");
521 got = returned.unclen;
522 bytes_written += returned.clen;
523 if (returned.unclen == returned.clen)
531 msg("EOT detected in last 2 tape records!\n");
532 msg("Use a longer tape, decrease the size estimate\n");
533 quit("or use no size estimate at all.\n");
536 (void) close(slaves[f].fd);
538 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
543 msg("Closing %s\n", tape);
548 while (rmtopen(tape, 0) < 0)
555 (void) close(tapefd);
557 while ((f = OPEN(tape, 0)) < 0)
563 if (eot_script && spcl.c_type != TS_END) {
564 msg("Launching %s\n", eot_script);
565 eot_code = system_command(eot_script, tape, tapeno);
567 if (eot_code != 0 && eot_code != 1) {
568 msg("Dump aborted by the end of tape script\n");
580 if (nexttape || Mflag || (eot_code == 0) )
583 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
584 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
586 while (!query("Is the new volume mounted and ready to go?"))
587 if (query("Do you want to abort?")) {
596 struct req *p, *q = NULL, *prev;
598 int i, size, savedtapea, got;
599 union u_spcl *ntb, *otb;
600 struct slave_results returned;
605 tslp = &slaves[SLAVES];
606 ntb = (union u_spcl *)tslp->tblock[1];
609 * Each of the N slaves should have requests that need to
610 * be replayed on the next tape. Use the extra slave buffers
611 * (slaves[SLAVES]) to construct request lists to be sent to
612 * each slave in turn.
614 for (i = 0; i < SLAVES; i++) {
616 otb = (union u_spcl *)slp->tblock;
619 * For each request in the current slave, copy it to tslp.
623 for (p = slp->req; p->count > 0; p += p->count) {
626 *ntb++ = *otb++; /* copy the datablock also */
631 quit("rollforward: protocol botch");
643 nextblock = tslp->tblock;
644 savedtapea = spcl.c_tapea;
645 spcl.c_tapea = slp->tapea;
647 spcl.c_tapea = savedtapea;
648 lastspclrec = savedtapea - 1;
650 size = (char *)ntb - (char *)q;
651 if (atomic_write( slp->fd, (char *)q, size) != size) {
652 perror(" DUMP: error writing command pipe");
657 lastfirstrec = slp->firstrec;
659 if (++slp >= &slaves[SLAVES])
664 if (prev->dblk != 0) {
666 * If the last one was a disk block, make the
667 * first of this one be the last bit of that disk
670 q->dblk = prev->dblk +
671 prev->count * (TP_BSIZE / DEV_BSIZE);
672 ntb = (union u_spcl *)tslp->tblock;
675 * It wasn't a disk block. Copy the data to its
676 * new location in the buffer.
679 *((union u_spcl *)tslp->tblock) = *ntb;
680 ntb = (union u_spcl *)tslp->tblock[1];
684 nextblock = slp->tblock;
687 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
688 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
695 * Clear the first slaves' response. One hopes that it
696 * worked ok, otherwise the tape is much too short!
699 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
700 != sizeof returned) {
701 perror(" DUMP: error reading command pipe in master");
704 got = returned.unclen;
705 bytes_written += returned.clen;
706 if (returned.clen == returned.unclen)
714 quit("EOT detected at start of the tape!\n");
720 if (spcl.c_type != TS_END) {
721 for (i = 0; i < spcl.c_count; i++)
722 if (spcl.c_addr[i] != 0)
726 slp->firstrec = lastfirstrec + ntrec;
727 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
729 asize += tenths + returned.clen / density;
730 blockswritten += ntrec;
731 blocksthisvol += ntrec;
736 * We implement taking and restoring checkpoints on the tape level.
737 * When each tape is opened, a new process is created by forking; this
738 * saves all of the necessary context in the parent. The child
739 * continues the dump; the parent waits around, saving the context.
740 * If the child returns X_REWRITE, then it had problems writing that tape;
741 * this causes the parent to fork again, duplicating the context, and
742 * everything continues as if nothing had happened.
745 startnewtape(int top)
756 sigaddset(&sigs, SIGINT);
757 sigprocmask(SIG_BLOCK, &sigs, NULL);
758 #else /* __linux__ */
760 void (*interrupt_save)();
762 sig_t interrupt_save;
764 interrupt_save = signal(SIGINT, SIG_IGN);
765 #endif /* __linux__ */
767 parentpid = getpid();
768 tapea_volume = spcl.c_tapea;
769 tapea_bytes = bytes_written;
770 tstart_volume = time(NULL);
774 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
776 (void)signal(SIGINT, interrupt_save);
779 * All signals are inherited...
783 msg("Context save fork fails in parent %d\n", parentpid);
789 * save the context by waiting
790 * until the child doing all of the work returns.
791 * don't catch the interrupt
794 sigprocmask(SIG_BLOCK, &sigs, NULL);
796 signal(SIGINT, SIG_IGN);
799 msg("Tape: %d; parent process: %d child process %d\n",
800 tapeno+1, parentpid, childpid);
802 while ((waitpid = wait(&status)) != childpid)
803 if (waitpid != rshpid)
804 msg("Parent %d waiting for child %d has another child %d return\n",
805 parentpid, childpid, waitpid);
807 msg("Child %d returns LOB status %o\n",
808 childpid, status&0xFF);
810 status = (status >> 8) & 0xFF;
814 msg("Child %d finishes X_FINOK\n", childpid);
817 msg("Child %d finishes X_ABORT\n", childpid);
820 msg("Child %d finishes X_REWRITE\n", childpid);
823 msg("Child %d finishes unknown %d\n",
834 goto restore_check_point;
836 msg("Bad return code from dump: %d\n", status);
840 } else { /* we are the child; just continue */
842 sleep(4); /* allow time for parent's message to get out */
843 msg("Child on Tape %d has parent %d, my pid = %d\n",
844 tapeno+1, parentpid, getpid());
847 * If we have a name like "/dev/rmt0,/dev/rmt1",
848 * use the name before the comma first, and save
849 * the remaining names for subsequent volumes.
851 tapeno++; /* current tape sequence */
853 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
854 tape[MAXPATHLEN - 1] = '\0';
855 msg("Dumping volume %d on %s\n", tapeno, tape);
857 else if (nexttape || strchr(tapeprefix, ',')) {
858 if (nexttape && *nexttape)
859 tapeprefix = nexttape;
860 if ((p = strchr(tapeprefix, ',')) != NULL) {
865 strncpy(tape, tapeprefix, MAXPATHLEN);
866 tape[MAXPATHLEN - 1] = '\0';
867 msg("Dumping volume %d on %s\n", tapeno, tape);
870 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
872 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
874 while ((tapefd = (pipeout ? fileno(stdout) :
875 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
878 msg("Cannot open output \"%s\".\n", tape);
879 if (!query("Do you want to retry the open?"))
886 struct mtget mt_stat;
887 magtapeout = ioctl(tapefd, MTIOCGET, (char *)&mt_stat) == 0;
889 msg("Output is to %s\n",
890 magtapeout ? "tape" : "file/pipe");
894 enslave(); /* Share open tape file descriptor with slaves */
899 newtape++; /* new tape signal */
900 spcl.c_count = slp->count;
902 * measure firstrec in TP_BSIZE units since restore doesn't
903 * know the correct ntrec value...
905 spcl.c_firstrec = slp->firstrec;
907 spcl.c_type = TS_TAPE;
908 spcl.c_flags |= DR_NEWHEADER;
909 spcl.c_ntrec = ntrec;
911 spcl.c_flags |= DR_COMPRESSED;
912 writeheader((dump_ino_t)slp->inode);
913 spcl.c_flags &=~ DR_NEWHEADER;
914 msg("Volume %d started with block %ld at: %s", tapeno,
915 spcl.c_tapea, ctime(&tstart_volume));
917 msg("Volume %d begins with blocks from inode %d\n",
919 if (tapeno < TP_NINOS)
920 volinfo[tapeno] = slp->inode;
928 if (master != 0 && master != getpid())
929 /* Signals master to call dumpabort */
930 (void) kill(master, SIGTERM);
933 msg("The ENTIRE dump is aborted.\n");
946 msg("pid = %d exits with status %d\n", getpid(), status);
952 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
958 siglongjmp(jmpbuf, 1);
966 #ifdef LINUX_FORK_BUG
974 { struct sigaction sa;
975 memset(&sa, 0, sizeof sa);
976 sigemptyset(&sa.sa_mask);
977 sa.sa_handler = dumpabort;
978 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
979 sa.sa_handler = sigpipe;
980 sigaction(SIGPIPE, &sa, NULL);
981 sa.sa_handler = proceed;
982 sa.sa_flags = SA_RESTART;
983 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
986 for (i = 0; i < SLAVES; i++) {
987 if (i == slp - &slaves[0]) {
993 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
994 (slaves[i].pid = fork()) < 0)
995 quit("too many slaves, %d (recompile smaller): %s\n",
998 slaves[i].fd = cmd[1];
1000 if (slaves[i].pid == 0) { /* Slave starts up here */
1002 for (j = 0; j <= i; j++)
1003 (void) close(slaves[j].fd);
1005 sigaddset(&sigs, SIGINT); /* Master handles this */
1006 #if defined(SIGINFO)
1007 sigaddset(&sigs, SIGINFO);
1009 sigprocmask(SIG_BLOCK, &sigs, NULL);
1011 #ifdef LINUX_FORK_BUG
1012 if (atomic_write( cmd[0], (char *) &i, sizeof i)
1014 quit("master/slave protocol botched 3\n");
1016 doslave(cmd[0], i, (slaves[i].pid == slp->pid));
1023 #ifdef LINUX_FORK_BUG
1025 * Wait for all slaves to _actually_ start to circumvent a bug in
1026 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1027 * returned from fork() causes a SEGV in the child process
1029 for (i = 0; i < SLAVES; i++)
1030 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
1031 quit("master/slave protocol botched 4\n");
1034 for (i = 0; i < SLAVES; i++)
1035 (void) atomic_write( slaves[i].fd,
1036 (char *) &slaves[(i + 1) % SLAVES].pid,
1037 sizeof slaves[0].pid);
1047 for (i = 0; i < SLAVES; i++)
1048 if (slaves[i].pid > 0) {
1049 (void) kill(slaves[i].pid, SIGKILL);
1055 * Synchronization - each process waits for a SIGUSR2 from the
1056 * previous process before writing to the tape, and sends SIGUSR2
1057 * to the next process when the tape write completes. On tape errors
1058 * a SIGUSR1 is sent to the master which then terminates all of the
1062 doslave(int cmd, int slave_number, int first)
1065 int nextslave, size, eot_count, bufsize;
1066 volatile int wrote = 0;
1068 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1069 struct tapebuf *comp_buf = NULL;
1070 int compresult, do_compress = !first;
1071 unsigned long worklen;
1073 unsigned int worklen2;
1075 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1076 struct slave_results returns;
1081 long long curtapepos;
1082 union u_spcl *uspclptr;
1083 struct s_spcl *spclptr;
1084 #endif /* USE_QFA */
1088 sigaddset(&set, SIGUSR2);
1089 sigprocmask(SIG_BLOCK, &set, NULL);
1093 * Need our own seek pointer.
1095 (void) close(diskfd);
1096 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1097 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1100 (void)ioctl(diskfd, BLKFLSBUF);
1103 retval = dump_fs_open(disk, &fs);
1105 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1106 #endif /* __linux__ */
1109 * Need the pid of the next slave in the loop...
1111 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1112 != sizeof nextslave) {
1113 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1116 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1117 /* if we're doing a compressed dump, allocate the compress buffer */
1119 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1120 if (comp_buf == NULL)
1121 quit("couldn't allocate a compress buffer.\n");
1123 comp_buf->flags = COMPRESS_BZLIB;
1125 comp_buf->flags = COMPRESS_ZLIB;
1127 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1130 * Get list of blocks to dump, read the blocks into tape buffer
1132 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1133 struct req *p = slp->req;
1135 for (trecno = 0; trecno < ntrec;
1136 trecno += p->count, p += p->count) {
1137 if (p->dblk) { /* read a disk block */
1138 bread(p->dblk, slp->tblock[trecno],
1139 p->count * TP_BSIZE);
1140 } else { /* read record from pipe */
1141 if (p->count != 1 || atomic_read( cmd,
1142 (char *)slp->tblock[trecno],
1143 TP_BSIZE) != TP_BSIZE)
1144 quit("master/slave protocol botched.\n");
1148 /* Try to write the data... */
1152 buffer = (char *) slp->tblock[0]; /* set write pointer */
1153 bufsize = writesize; /* length to write */
1154 returns.clen = returns.unclen = bufsize;
1156 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1158 * When writing a compressed dump, each block except
1159 * the first one on each tape is written
1160 * from struct tapebuf with an 4 byte prefix
1161 * followed by the data. This can be less than
1162 * writesize. Restore, on a short read, can compare the
1163 * length read to the compressed length in the header
1164 * to verify that the read was good. Blocks which don't
1165 * compress well are written uncompressed.
1166 * The first block written by each slave is not compressed
1167 * and does not have a prefix.
1170 if (compressed && do_compress) {
1171 comp_buf->length = bufsize;
1172 worklen = TP_BSIZE + writesize;
1176 compresult = compress2(comp_buf->buf,
1178 (char *)slp->tblock[0],
1181 if (compresult == Z_OK)
1186 #endif /* HAVE_ZLIB */
1190 compresult = BZ2_bzBuffToBuffCompress(
1193 (char *)slp->tblock[0],
1198 if (compresult == BZ_OK)
1204 #endif /* HAVE_BZLIB */
1205 if (compresult && worklen <= (writesize - 16)) {
1206 /* write the compressed buffer */
1207 comp_buf->length = worklen;
1208 comp_buf->compressed = 1;
1209 buffer = (char *) comp_buf;
1210 returns.clen = bufsize = worklen + sizeof(struct tapebuf);
1213 /* write the data uncompressed */
1214 comp_buf->length = writesize;
1215 comp_buf->compressed = 0;
1216 buffer = (char *) comp_buf;
1217 returns.clen = bufsize = writesize + sizeof(struct tapebuf);
1218 returns.unclen = returns.clen;
1219 memcpy(comp_buf->buf, (char *)slp->tblock[0], writesize);
1222 /* compress the remaining blocks if we're compressing */
1223 do_compress = compressed;
1224 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1226 if (sigsetjmp(jmpbuf, 1) == 0) {
1235 if (gTapeposfd >= 0) {
1237 for (i = 0; i < ntrec; ++i) {
1238 uspclptr = (union u_spcl *)&slp->tblock[i];
1239 spclptr = &uspclptr->s_spcl;
1240 if ((spclptr->c_magic == NFS_MAGIC) &&
1241 (spclptr->c_type == TS_INODE) &&
1242 ((spclptr->c_dinode.di_mode & S_IFMT) != IFDIR) &&
1243 (spclptr->c_date == gThisDumpDate)) {
1244 /* if an error occured previously don't
1247 if ((gtperr = GetTapePos(&curtapepos)) == 0)
1248 MkTapeString(spclptr, curtapepos);
1253 #endif /* USE_QFA */
1255 while (eot_count < 10 && size < bufsize) {
1258 wrote = rmtwrite(buffer + size, bufsize - size);
1261 wrote = write(tapefd, buffer + size, bufsize - size);
1263 printf("slave %d wrote %d\n", slave_number, wrote);
1273 if (size != bufsize)
1274 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1275 slave_number, size, bufsize);
1279 * Handle ENOSPC as an EOT condition.
1281 if (wrote < 0 && errno == ENOSPC) {
1287 returns.clen = returns.unclen = 0;
1290 * pass errno back to master for special handling
1293 returns.unclen = -errno;
1296 * pass size of data and size of write back to master
1297 * (for EOT handling)
1299 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1302 * Signal the next slave to go.
1304 (void) kill(nextslave, SIGUSR2);
1307 quit("error reading command pipe: %s\n", strerror(errno));
1311 * Since a read from a pipe may not return all we asked for,
1312 * or a write may not write all we ask if we get a signal,
1313 * loop until the count is satisfied (or error).
1316 atomic_read(int fd, void *buf, size_t count)
1318 int got, need = count;
1321 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1323 } while (got == -1 && errno == EINTR);
1324 return (got < 0 ? got : count - need);
1328 * Since a read from a pipe may not return all we asked for,
1329 * or a write may not write all we ask if we get a signal,
1330 * loop until the count is satisfied (or error).
1333 atomic_write(int fd, const void *buf, size_t count)
1335 int got, need = count;
1338 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1340 } while (got == -1 && errno == EINTR);
1341 return (got < 0 ? got : count - need);
1347 * read the current tape position
1350 GetTapePos(long long *pos)
1356 *pos = (long long) rmtseek(0, SEEK_CUR);
1365 err = (ioctl(tapefd, MTIOCPOS, &mtpos) < 0);
1366 *pos = (long long)mtpos;
1369 *pos = LSEEK(tapefd, 0, SEEK_CUR);
1375 msg("[%ld] error: %d (getting tapepos: %lld)\n", getpid(),
1383 MkTapeString(struct s_spcl *spclptr, long long curtapepos) {
1386 msg("inode %ld at tapepos %lld\n", spclptr->c_inumber, curtapepos);
1389 snprintf(gTps, sizeof(gTps), "%ld\t%d\t%lld\n",
1390 (unsigned long)spclptr->c_inumber,
1393 gTps[sizeof(gTps) - 1] = '\0';
1394 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1395 warn("error writing tapepos file.\n");
1398 #endif /* USE_QFA */