2 * Copyright (c) 1980, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)tape.c 8.4 (Berkeley) 5/1/95
34 * $FreeBSD: src/sbin/dump/tape.c,v 1.12.2.3 2002/02/23 22:32:51 iedowse Exp $
35 * $DragonFly: src/sbin/dump/tape.c,v 1.3 2003/08/08 04:18:37 dillon Exp $
38 #include <sys/param.h>
39 #include <sys/socket.h>
44 #include <sys/vnode.h>
47 #include <ufs/inode.h>
49 #include <vfs/ufs/dinode.h>
50 #include <vfs/ufs/fs.h>
53 #include <protocols/dumprestore.h>
70 int writesize; /* size of malloc()ed buffer for tape */
71 long lastspclrec = -1; /* tape block number of last written header */
72 int trecno = 0; /* next record to write in current block */
73 extern long blocksperfile; /* number of blocks per output file */
74 long blocksthisvol; /* number of blocks on current output file */
75 extern int ntrec; /* blocking factor on tape */
80 static int atomic __P((ssize_t (*)(), int, char *, int));
81 static void doslave __P((int, int));
82 static void enslave __P((void));
83 static void flushtape __P((void));
84 static void killall __P((void));
85 static void rollforward __P((void));
88 * Concurrent dump mods (Caltech) - disk block reading and tape writing
89 * are exported to several slave processes. While one slave writes the
90 * tape, the others read disk blocks; they pass control of the tape in
91 * a ring via signals. The parent process traverses the filesystem and
92 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
93 * The following structure defines the instruction packets sent to slaves.
101 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
103 int tapea; /* header number at start of this chunk */
104 int count; /* count to next header (used for TS_TAPE */
106 int inode; /* inode that we are currently dealing with */
107 int fd; /* FD for this slave */
108 int pid; /* PID for this slave */
109 int sent; /* 1 == we've sent this slave requests */
110 int firstrec; /* record number of this block */
111 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
112 struct req *req; /* buffer for requests */
116 char (*nextblock)[TP_BSIZE];
118 int master; /* pid of master, for sending error signals */
119 int tenths; /* length of tape used per block written */
120 static int caught; /* have we caught the signal to proceed? */
121 static int ready; /* have we reached the lock point without having */
122 /* received the SIGUSR2 signal from the prev slave? */
123 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */
124 /* SIGUSR2 arrives from the previous slave */
129 int pgoff = getpagesize() - 1;
133 writesize = ntrec * TP_BSIZE;
134 reqsiz = (ntrec + 1) * sizeof(struct req);
136 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
137 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
138 * repositioning after stopping, i.e, streaming mode, where the gap is
139 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
141 if (blocksperfile == 0 && !unlimited)
142 tenths = writesize / density +
143 (cartridge ? 16 : density == 625 ? 5 : 8);
145 * Allocate tape buffer contiguous with the array of instruction
146 * packets, so flushtape() can write them together with one write().
147 * Align tape buffer on page boundary to speed up tape write().
149 for (i = 0; i <= SLAVES; i++) {
151 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
154 slaves[i].tblock = (char (*)[TP_BSIZE])
155 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
156 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
162 nextblock = slp->tblock;
172 slp->req[trecno].dblk = (daddr_t)0;
173 slp->req[trecno].count = 1;
175 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp;
177 bcopy(dp, *(nextblock)++, sizeof (union u_spcl));
180 lastspclrec = spcl.c_tapea;
188 dumpblock(blkno, size)
192 int avail, tpblks, dblkno;
194 dblkno = fsbtodb(sblock, blkno);
195 tpblks = size >> tp_bshift;
196 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
197 slp->req[trecno].dblk = dblkno;
198 slp->req[trecno].count = avail;
200 spcl.c_tapea += avail;
203 dblkno += avail << (tp_bshift - dev_bshift);
216 msg("write error on %s\n", tape);
217 quit("Cannot recover\n");
220 msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno);
221 broadcast("DUMP WRITE ERROR!\n");
222 if (!query("Do you want to restart?"))
224 msg("Closing this volume. Prepare to restart with new media;\n");
225 msg("this dump volume will be rewritten.\n");
237 quit("Broken pipe\n");
246 int siz = (char *)nextblock - (char *)slp->req;
248 slp->req[trecno].count = 0; /* Sentinel */
250 if (atomic(write, slp->fd, (char *)slp->req, siz) != siz)
251 quit("error writing command pipe: %s\n", strerror(errno));
252 slp->sent = 1; /* we sent a request, read the response later */
254 lastfirstrec = slp->firstrec;
256 if (++slp >= &slaves[SLAVES])
259 /* Read results back from next slave */
261 if (atomic(read, slp->fd, (char *)&got, sizeof got)
263 perror(" DUMP: error reading command pipe in master");
268 /* Check for end of tape */
269 if (got < writesize) {
270 msg("End of tape detected\n");
273 * Drain the results, don't care what the values were.
274 * If we read them here then trewind won't...
276 for (i = 0; i < SLAVES; i++) {
277 if (slaves[i].sent) {
278 if (atomic(read, slaves[i].fd,
279 (char *)&got, sizeof got)
281 perror(" DUMP: error reading command pipe in master");
295 if (spcl.c_type != TS_END) {
296 for (i = 0; i < spcl.c_count; i++)
297 if (spcl.c_addr[i] != 0)
300 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
301 slp->tapea = spcl.c_tapea;
302 slp->firstrec = lastfirstrec + ntrec;
304 nextblock = slp->tblock;
307 blockswritten += ntrec;
308 blocksthisvol += ntrec;
309 if (!pipeout && !unlimited && (blocksperfile ?
310 (blocksthisvol >= blocksperfile) : (asize > tsize))) {
324 for (f = 0; f < SLAVES; f++) {
326 * Drain the results, but unlike EOT we DO (or should) care
327 * what the return values were, since if we detect EOT after
328 * we think we've written the last blocks to the tape anyway,
329 * we have to replay those blocks with rollforward.
331 * fixme: punt for now.
333 if (slaves[f].sent) {
334 if (atomic(read, slaves[f].fd, (char *)&got, sizeof got)
336 perror(" DUMP: error reading command pipe in master");
340 if (got != writesize) {
341 msg("EOT detected in last 2 tape records!\n");
342 msg("Use a longer tape, decrease the size estimate\n");
343 quit("or use no size estimate at all.\n");
346 (void) close(slaves[f].fd);
348 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
354 msg("Closing %s\n", tape);
359 while (rmtopen(tape, 0) < 0)
365 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) {
369 (void) close(tapefd);
370 while ((f = open(tape, 0)) < 0)
378 time_t tstart_changevol, tend_changevol;
383 (void)time((time_t *)&(tstart_changevol));
385 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
386 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
388 while (!query("Is the new volume mounted and ready to go?"))
389 if (query("Do you want to abort?")) {
393 (void)time((time_t *)&(tend_changevol));
394 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1))
395 tstart_writing += (tend_changevol - tstart_changevol);
401 register struct req *p, *q, *prev;
402 register struct slave *tslp;
403 int i, size, savedtapea, got;
404 union u_spcl *ntb, *otb;
405 tslp = &slaves[SLAVES];
406 ntb = (union u_spcl *)tslp->tblock[1];
409 * Each of the N slaves should have requests that need to
410 * be replayed on the next tape. Use the extra slave buffers
411 * (slaves[SLAVES]) to construct request lists to be sent to
412 * each slave in turn.
414 for (i = 0; i < SLAVES; i++) {
416 otb = (union u_spcl *)slp->tblock;
419 * For each request in the current slave, copy it to tslp.
423 for (p = slp->req; p->count > 0; p += p->count) {
426 *ntb++ = *otb++; /* copy the datablock also */
431 quit("rollforward: protocol botch");
443 nextblock = tslp->tblock;
444 savedtapea = spcl.c_tapea;
445 spcl.c_tapea = slp->tapea;
447 spcl.c_tapea = savedtapea;
448 lastspclrec = savedtapea - 1;
450 size = (char *)ntb - (char *)q;
451 if (atomic(write, slp->fd, (char *)q, size) != size) {
452 perror(" DUMP: error writing command pipe");
456 if (++slp >= &slaves[SLAVES])
461 if (prev->dblk != 0) {
463 * If the last one was a disk block, make the
464 * first of this one be the last bit of that disk
467 q->dblk = prev->dblk +
468 prev->count * (TP_BSIZE / DEV_BSIZE);
469 ntb = (union u_spcl *)tslp->tblock;
472 * It wasn't a disk block. Copy the data to its
473 * new location in the buffer.
476 *((union u_spcl *)tslp->tblock) = *ntb;
477 ntb = (union u_spcl *)tslp->tblock[1];
481 nextblock = slp->tblock;
487 * Clear the first slaves' response. One hopes that it
488 * worked ok, otherwise the tape is much too short!
491 if (atomic(read, slp->fd, (char *)&got, sizeof got)
493 perror(" DUMP: error reading command pipe in master");
498 if (got != writesize) {
499 quit("EOT detected at start of the tape!\n");
505 * We implement taking and restoring checkpoints on the tape level.
506 * When each tape is opened, a new process is created by forking; this
507 * saves all of the necessary context in the parent. The child
508 * continues the dump; the parent waits around, saving the context.
509 * If the child returns X_REWRITE, then it had problems writing that tape;
510 * this causes the parent to fork again, duplicating the context, and
511 * everything continues as if nothing had happened.
523 void (*interrupt_save)();
525 sig_t interrupt_save;
528 interrupt_save = signal(SIGINT, SIG_IGN);
529 parentpid = getpid();
532 (void)signal(SIGINT, interrupt_save);
534 * All signals are inherited...
536 setproctitle(NULL); /* Restore the proctitle. */
539 msg("Context save fork fails in parent %d\n", parentpid);
545 * save the context by waiting
546 * until the child doing all of the work returns.
547 * don't catch the interrupt
549 signal(SIGINT, SIG_IGN);
551 msg("Tape: %d; parent process: %d child process %d\n",
552 tapeno+1, parentpid, childpid);
554 while ((waitpid = wait(&status)) != childpid)
555 msg("Parent %d waiting for child %d has another child %d return\n",
556 parentpid, childpid, waitpid);
558 msg("Child %d returns LOB status %o\n",
559 childpid, status&0xFF);
561 status = (status >> 8) & 0xFF;
565 msg("Child %d finishes X_FINOK\n", childpid);
568 msg("Child %d finishes X_ABORT\n", childpid);
571 msg("Child %d finishes X_REWRITE\n", childpid);
574 msg("Child %d finishes unknown %d\n",
585 goto restore_check_point;
587 msg("Bad return code from dump: %d\n", status);
591 } else { /* we are the child; just continue */
593 sleep(4); /* allow time for parent's message to get out */
594 msg("Child on Tape %d has parent %d, my pid = %d\n",
595 tapeno+1, parentpid, getpid());
598 * If we have a name like "/dev/rmt0,/dev/rmt1",
599 * use the name before the comma first, and save
600 * the remaining names for subsequent volumes.
602 tapeno++; /* current tape sequence */
603 if (nexttape || strchr(tape, ',')) {
604 if (nexttape && *nexttape)
606 if ((p = strchr(tape, ',')) != NULL) {
611 msg("Dumping volume %d on %s\n", tapeno, tape);
614 while ((tapefd = (host ? rmtopen(tape, 2) :
615 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
617 while ((tapefd = (pipeout ? 1 :
618 open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
621 msg("Cannot open output \"%s\".\n", tape);
622 if (!query("Do you want to retry the open?"))
626 enslave(); /* Share open tape file descriptor with slaves */
627 signal(SIGINFO, infosch);
632 newtape++; /* new tape signal */
633 spcl.c_count = slp->count;
635 * measure firstrec in TP_BSIZE units since restore doesn't
636 * know the correct ntrec value...
638 spcl.c_firstrec = slp->firstrec;
640 spcl.c_type = TS_TAPE;
641 spcl.c_flags |= DR_NEWHEADER;
642 writeheader((ino_t)slp->inode);
643 spcl.c_flags &=~ DR_NEWHEADER;
645 msg("Volume %d begins with blocks from inode %d\n",
655 if (master != 0 && master != getpid())
656 /* Signals master to call dumpabort */
657 (void) kill(master, SIGTERM);
660 msg("The ENTIRE dump is aborted.\n");
674 msg("pid = %d exits with status %d\n", getpid(), status);
680 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
700 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */
701 signal(SIGPIPE, sigpipe);
702 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */
703 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */
705 for (i = 0; i < SLAVES; i++) {
706 if (i == slp - &slaves[0]) {
712 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
713 (slaves[i].pid = fork()) < 0)
714 quit("too many slaves, %d (recompile smaller): %s\n",
717 slaves[i].fd = cmd[1];
719 if (slaves[i].pid == 0) { /* Slave starts up here */
720 for (j = 0; j <= i; j++)
721 (void) close(slaves[j].fd);
722 signal(SIGINT, SIG_IGN); /* Master handles this */
728 for (i = 0; i < SLAVES; i++)
729 (void) atomic(write, slaves[i].fd,
730 (char *) &slaves[(i + 1) % SLAVES].pid,
731 sizeof slaves[0].pid);
741 for (i = 0; i < SLAVES; i++)
742 if (slaves[i].pid > 0) {
743 (void) kill(slaves[i].pid, SIGKILL);
749 * Synchronization - each process has a lockfile, and shares file
750 * descriptors to the following process's lockfile. When our write
751 * completes, we release our lock on the following process's lock-
752 * file, allowing the following process to lock it and proceed. We
753 * get the lock back for the next cycle by swapping descriptors.
756 doslave(cmd, slave_number)
761 int nextslave, size, wrote, eot_count;
764 * Need our own seek pointer.
766 (void) close(diskfd);
767 if ((diskfd = open(disk, O_RDONLY)) < 0)
768 quit("slave couldn't reopen disk: %s\n", strerror(errno));
771 * Need the pid of the next slave in the loop...
773 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave))
774 != sizeof nextslave) {
775 quit("master/slave protocol botched - didn't get pid of next slave.\n");
779 * Get list of blocks to dump, read the blocks into tape buffer
781 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) {
782 register struct req *p = slp->req;
784 for (trecno = 0; trecno < ntrec;
785 trecno += p->count, p += p->count) {
787 bread(p->dblk, slp->tblock[trecno],
788 p->count * TP_BSIZE);
790 if (p->count != 1 || atomic(read, cmd,
791 (char *)slp->tblock[trecno],
792 TP_BSIZE) != TP_BSIZE)
793 quit("master/slave protocol botched.\n");
796 if (setjmp(jmpbuf) == 0) {
804 /* Try to write the data... */
808 while (eot_count < 10 && size < writesize) {
811 wrote = rmtwrite(slp->tblock[0]+size,
815 wrote = write(tapefd, slp->tblock[0]+size,
818 printf("slave %d wrote %d\n", slave_number, wrote);
828 if (size != writesize)
829 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
830 slave_number, size, writesize);
834 * Handle ENOSPC as an EOT condition.
836 if (wrote < 0 && errno == ENOSPC) {
845 (void) kill(master, SIGUSR1);
850 * pass size of write back to master
853 (void) atomic(write, cmd, (char *)&size, sizeof size);
857 * If partial write, don't want next slave to go.
858 * Also jolts him awake.
860 (void) kill(nextslave, SIGUSR2);
863 quit("error reading command pipe: %s\n", strerror(errno));
867 * Since a read from a pipe may not return all we asked for,
868 * or a write may not write all we ask if we get a signal,
869 * loop until the count is satisfied (or error).
872 atomic(func, fd, buf, count)
878 int got, need = count;
880 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0)
882 return (got < 0 ? got : count - need);