2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1980, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95";
36 static const char rcsid[] =
40 #include <sys/param.h>
41 #include <sys/socket.h>
45 #include <ufs/ufs/dinode.h>
46 #include <ufs/ffs/fs.h>
48 #include <protocols/dumprestore.h>
63 int writesize; /* size of malloc()ed buffer for tape */
64 int64_t lastspclrec = -1; /* tape block number of last written header */
65 int trecno = 0; /* next record to write in current block */
66 extern long blocksperfile; /* number of blocks per output file */
67 long blocksthisvol; /* number of blocks on current output file */
68 extern int ntrec; /* blocking factor on tape */
74 static int atomic(ssize_t (*)(), int, char *, int);
75 static void doslave(int, int);
76 static void enslave(void);
77 static void flushtape(void);
78 static void killall(void);
79 static void rollforward(void);
82 * Concurrent dump mods (Caltech) - disk block reading and tape writing
83 * are exported to several slave processes. While one slave writes the
84 * tape, the others read disk blocks; they pass control of the tape in
85 * a ring via signals. The parent process traverses the file system and
86 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
87 * The following structure defines the instruction packets sent to slaves.
95 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
97 int64_t tapea; /* header number at start of this chunk */
98 int64_t firstrec; /* record number of this block */
99 int count; /* count to next header (used for TS_TAPE */
101 int inode; /* inode that we are currently dealing with */
102 int fd; /* FD for this slave */
103 int pid; /* PID for this slave */
104 int sent; /* 1 == we've sent this slave requests */
105 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
106 struct req *req; /* buffer for requests */
110 char (*nextblock)[TP_BSIZE];
112 int master; /* pid of master, for sending error signals */
113 int tenths; /* length of tape used per block written */
114 static volatile sig_atomic_t caught; /* have we caught the signal to proceed? */
115 static volatile sig_atomic_t ready; /* reached the lock point without having */
116 /* received the SIGUSR2 signal from the prev slave? */
117 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */
118 /* SIGUSR2 arrives from the previous slave */
123 int pgoff = getpagesize() - 1;
127 writesize = ntrec * TP_BSIZE;
128 reqsiz = (ntrec + 1) * sizeof(struct req);
130 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
131 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
132 * repositioning after stopping, i.e, streaming mode, where the gap is
133 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
135 if (blocksperfile == 0 && !unlimited)
136 tenths = writesize / density +
137 (cartridge ? 16 : density == 625 ? 5 : 8);
139 * Allocate tape buffer contiguous with the array of instruction
140 * packets, so flushtape() can write them together with one write().
141 * Align tape buffer on page boundary to speed up tape write().
143 for (i = 0; i <= SLAVES; i++) {
145 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
148 slaves[i].tblock = (char (*)[TP_BSIZE])
149 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
150 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
156 nextblock = slp->tblock;
161 writerec(char *dp, int isspcl)
164 slp->req[trecno].dblk = (ufs2_daddr_t)0;
165 slp->req[trecno].count = 1;
166 /* Can't do a structure assignment due to alignment problems */
167 bcopy(dp, *(nextblock)++, sizeof (union u_spcl));
169 lastspclrec = spcl.c_tapea;
177 dumpblock(ufs2_daddr_t blkno, int size)
182 dblkno = fsbtodb(sblock, blkno);
183 tpblks = size >> tp_bshift;
184 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
185 slp->req[trecno].dblk = dblkno;
186 slp->req[trecno].count = avail;
188 spcl.c_tapea += avail;
191 dblkno += avail << (tp_bshift - dev_bshift);
199 tperror(int signo __unused)
203 msg("write error on %s\n", tape);
204 quit("Cannot recover\n");
207 msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno);
208 broadcast("DUMP WRITE ERROR!\n");
209 if (!query("Do you want to restart?"))
211 msg("Closing this volume. Prepare to restart with new media;\n");
212 msg("this dump volume will be rewritten.\n");
220 sigpipe(int signo __unused)
223 quit("Broken pipe\n");
230 int64_t lastfirstrec;
232 int siz = (char *)nextblock - (char *)slp->req;
234 slp->req[trecno].count = 0; /* Sentinel */
236 if (atomic(write, slp->fd, (char *)slp->req, siz) != siz)
237 quit("error writing command pipe: %s\n", strerror(errno));
238 slp->sent = 1; /* we sent a request, read the response later */
240 lastfirstrec = slp->firstrec;
242 if (++slp >= &slaves[SLAVES])
245 /* Read results back from next slave */
247 if (atomic(read, slp->fd, (char *)&got, sizeof got)
249 perror(" DUMP: error reading command pipe in master");
254 /* Check for end of tape */
255 if (got < writesize) {
256 msg("End of tape detected\n");
259 * Drain the results, don't care what the values were.
260 * If we read them here then trewind won't...
262 for (i = 0; i < SLAVES; i++) {
263 if (slaves[i].sent) {
264 if (atomic(read, slaves[i].fd,
265 (char *)&got, sizeof got)
267 perror(" DUMP: error reading command pipe in master");
281 if (spcl.c_type != TS_END) {
282 for (i = 0; i < spcl.c_count; i++)
283 if (spcl.c_addr[i] != 0)
286 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
287 slp->tapea = spcl.c_tapea;
288 slp->firstrec = lastfirstrec + ntrec;
290 nextblock = slp->tblock;
293 blockswritten += ntrec;
294 blocksthisvol += ntrec;
295 if (!pipeout && !unlimited && (blocksperfile ?
296 (blocksthisvol >= blocksperfile) : (asize > tsize))) {
310 for (f = 0; f < SLAVES; f++) {
312 * Drain the results, but unlike EOT we DO (or should) care
313 * what the return values were, since if we detect EOT after
314 * we think we've written the last blocks to the tape anyway,
315 * we have to replay those blocks with rollforward.
317 * fixme: punt for now.
319 if (slaves[f].sent) {
320 if (atomic(read, slaves[f].fd, (char *)&got, sizeof got)
322 perror(" DUMP: error reading command pipe in master");
326 if (got != writesize) {
327 msg("EOT detected in last 2 tape records!\n");
328 msg("Use a longer tape, decrease the size estimate\n");
329 quit("or use no size estimate at all.\n");
332 (void) close(slaves[f].fd);
334 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
340 msg("Closing %s\n", tape);
344 (void)pclose(popenfp);
351 while (rmtopen(tape, 0) < 0)
357 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) {
361 (void) close(tapefd);
362 while ((f = open(tape, 0)) < 0)
370 time_t tstart_changevol, tend_changevol;
375 (void)time((time_t *)&(tstart_changevol));
377 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
378 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
380 while (!query("Is the new volume mounted and ready to go?"))
381 if (query("Do you want to abort?")) {
385 (void)time((time_t *)&(tend_changevol));
386 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1))
387 tstart_writing += (tend_changevol - tstart_changevol);
393 struct req *p, *q, *prev;
397 union u_spcl *ntb, *otb;
398 tslp = &slaves[SLAVES];
399 ntb = (union u_spcl *)tslp->tblock[1];
402 * Each of the N slaves should have requests that need to
403 * be replayed on the next tape. Use the extra slave buffers
404 * (slaves[SLAVES]) to construct request lists to be sent to
405 * each slave in turn.
407 for (i = 0; i < SLAVES; i++) {
409 otb = (union u_spcl *)slp->tblock;
412 * For each request in the current slave, copy it to tslp.
416 for (p = slp->req; p->count > 0; p += p->count) {
419 *ntb++ = *otb++; /* copy the datablock also */
424 quit("rollforward: protocol botch");
436 nextblock = tslp->tblock;
437 savedtapea = spcl.c_tapea;
438 spcl.c_tapea = slp->tapea;
440 spcl.c_tapea = savedtapea;
441 lastspclrec = savedtapea - 1;
443 size = (char *)ntb - (char *)q;
444 if (atomic(write, slp->fd, (char *)q, size) != size) {
445 perror(" DUMP: error writing command pipe");
449 if (++slp >= &slaves[SLAVES])
454 if (prev->dblk != 0) {
456 * If the last one was a disk block, make the
457 * first of this one be the last bit of that disk
460 q->dblk = prev->dblk +
461 prev->count * (TP_BSIZE / DEV_BSIZE);
462 ntb = (union u_spcl *)tslp->tblock;
465 * It wasn't a disk block. Copy the data to its
466 * new location in the buffer.
469 *((union u_spcl *)tslp->tblock) = *ntb;
470 ntb = (union u_spcl *)tslp->tblock[1];
474 nextblock = slp->tblock;
480 * Clear the first slaves' response. One hopes that it
481 * worked ok, otherwise the tape is much too short!
484 if (atomic(read, slp->fd, (char *)&got, sizeof got)
486 perror(" DUMP: error reading command pipe in master");
491 if (got != writesize) {
492 quit("EOT detected at start of the tape!\n");
498 * We implement taking and restoring checkpoints on the tape level.
499 * When each tape is opened, a new process is created by forking; this
500 * saves all of the necessary context in the parent. The child
501 * continues the dump; the parent waits around, saving the context.
502 * If the child returns X_REWRITE, then it had problems writing that tape;
503 * this causes the parent to fork again, duplicating the context, and
504 * everything continues as if nothing had happened.
507 startnewtape(int top)
513 sig_t interrupt_save;
515 interrupt_save = signal(SIGINT, SIG_IGN);
516 parentpid = getpid();
519 (void)signal(SIGINT, interrupt_save);
521 * All signals are inherited...
523 setproctitle(NULL); /* Restore the proctitle. */
526 msg("Context save fork fails in parent %d\n", parentpid);
532 * save the context by waiting
533 * until the child doing all of the work returns.
534 * don't catch the interrupt
536 signal(SIGINT, SIG_IGN);
538 msg("Tape: %d; parent process: %d child process %d\n",
539 tapeno+1, parentpid, childpid);
541 if (waitpid(childpid, &status, 0) == -1)
542 msg("Waiting for child %d: %s\n", childpid,
545 msg("Child %d returns LOB status %o\n",
546 childpid, status&0xFF);
548 status = (status >> 8) & 0xFF;
552 msg("Child %d finishes X_FINOK\n", childpid);
555 msg("Child %d finishes X_ABORT\n", childpid);
558 msg("Child %d finishes X_REWRITE\n", childpid);
561 msg("Child %d finishes unknown %d\n",
572 goto restore_check_point;
574 msg("Bad return code from dump: %d\n", status);
578 } else { /* we are the child; just continue */
580 sleep(4); /* allow time for parent's message to get out */
581 msg("Child on Tape %d has parent %d, my pid = %d\n",
582 tapeno+1, parentpid, getpid());
585 * If we have a name like "/dev/rmt0,/dev/rmt1",
586 * use the name before the comma first, and save
587 * the remaining names for subsequent volumes.
589 tapeno++; /* current tape sequence */
590 if (nexttape || strchr(tape, ',')) {
591 if (nexttape && *nexttape)
593 if ((p = strchr(tape, ',')) != NULL) {
598 msg("Dumping volume %d on %s\n", tapeno, tape);
601 tapefd = STDOUT_FILENO;
602 } else if (popenout) {
603 char volno[sizeof("2147483647")];
605 (void)sprintf(volno, "%d", spcl.c_volume + 1);
606 if (setenv("DUMP_VOLUME", volno, 1) == -1) {
607 msg("Cannot set $DUMP_VOLUME.\n");
610 popenfp = popen(popenout, "w");
611 if (popenfp == NULL) {
612 msg("Cannot open output pipeline \"%s\".\n",
616 tapefd = fileno(popenfp);
619 while ((tapefd = (host ? rmtopen(tape, 2) :
620 open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
623 open(tape, O_WRONLY|O_CREAT, 0666)) < 0)
626 msg("Cannot open output \"%s\".\n", tape);
627 if (!query("Do you want to retry the open?"))
632 enslave(); /* Share open tape file descriptor with slaves */
634 close(tapefd); /* Give up our copy of it. */
635 signal(SIGINFO, infosch);
640 newtape++; /* new tape signal */
641 spcl.c_count = slp->count;
643 * measure firstrec in TP_BSIZE units since restore doesn't
644 * know the correct ntrec value...
646 spcl.c_firstrec = slp->firstrec;
648 spcl.c_type = TS_TAPE;
649 writeheader((ino_t)slp->inode);
651 msg("Volume %d begins with blocks from inode %d\n",
657 dumpabort(int signo __unused)
660 if (master != 0 && master != getpid())
661 /* Signals master to call dumpabort */
662 (void) kill(master, SIGTERM);
665 msg("The ENTIRE dump is aborted.\n");
679 msg("pid = %d exits with status %d\n", getpid(), status);
685 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
688 proceed(int signo __unused)
704 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */
705 signal(SIGPIPE, sigpipe);
706 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */
707 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */
709 for (i = 0; i < SLAVES; i++) {
710 if (i == slp - &slaves[0]) {
716 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
717 (slaves[i].pid = fork()) < 0)
718 quit("too many slaves, %d (recompile smaller): %s\n",
721 slaves[i].fd = cmd[1];
723 if (slaves[i].pid == 0) { /* Slave starts up here */
724 for (j = 0; j <= i; j++)
725 (void) close(slaves[j].fd);
726 signal(SIGINT, SIG_IGN); /* Master handles this */
732 for (i = 0; i < SLAVES; i++)
733 (void) atomic(write, slaves[i].fd,
734 (char *) &slaves[(i + 1) % SLAVES].pid,
735 sizeof slaves[0].pid);
745 for (i = 0; i < SLAVES; i++)
746 if (slaves[i].pid > 0) {
747 (void) kill(slaves[i].pid, SIGKILL);
753 * Synchronization - each process has a lockfile, and shares file
754 * descriptors to the following process's lockfile. When our write
755 * completes, we release our lock on the following process's lock-
756 * file, allowing the following process to lock it and proceed. We
757 * get the lock back for the next cycle by swapping descriptors.
760 doslave(int cmd, int slave_number)
763 int nextslave, size, wrote, eot_count;
766 * Need our own seek pointer.
768 (void) close(diskfd);
769 if ((diskfd = open(disk, O_RDONLY)) < 0)
770 quit("slave couldn't reopen disk: %s\n", strerror(errno));
773 * Need the pid of the next slave in the loop...
775 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave))
776 != sizeof nextslave) {
777 quit("master/slave protocol botched - didn't get pid of next slave.\n");
781 * Get list of blocks to dump, read the blocks into tape buffer
783 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) {
784 struct req *p = slp->req;
786 for (trecno = 0; trecno < ntrec;
787 trecno += p->count, p += p->count) {
789 blkread(p->dblk, slp->tblock[trecno],
790 p->count * TP_BSIZE);
792 if (p->count != 1 || atomic(read, cmd,
793 (char *)slp->tblock[trecno],
794 TP_BSIZE) != TP_BSIZE)
795 quit("master/slave protocol botched.\n");
798 if (setjmp(jmpbuf) == 0) {
806 /* Try to write the data... */
811 while (eot_count < 10 && size < writesize) {
814 wrote = rmtwrite(slp->tblock[0]+size,
818 wrote = write(tapefd, slp->tblock[0]+size,
821 printf("slave %d wrote %d\n", slave_number, wrote);
831 if (size != writesize)
832 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
833 slave_number, size, writesize);
837 * Handle ENOSPC as an EOT condition.
839 if (wrote < 0 && errno == ENOSPC) {
848 (void) kill(master, SIGUSR1);
853 * pass size of write back to master
856 (void) atomic(write, cmd, (char *)&size, sizeof size);
860 * If partial write, don't want next slave to go.
861 * Also jolts him awake.
863 (void) kill(nextslave, SIGUSR2);
866 quit("error reading command pipe: %s\n", strerror(errno));
870 * Since a read from a pipe may not return all we asked for,
871 * or a write may not write all we ask if we get a signal,
872 * loop until the count is satisfied (or error).
875 atomic(ssize_t (*func)(), int fd, char *buf, int count)
877 int got, need = count;
879 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0)
881 return (got < 0 ? got : count - need);