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 1.11 2005/04/13 14:32:01 joerg 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_read(int, void *, int);
81 static int atomic_write(int, const void *, int);
83 static void doslave(int, int);
85 static void doslave(int);
87 static void enslave(void);
88 static void flushtape(void);
89 static void killall(void);
90 static void rollforward(void);
93 * Concurrent dump mods (Caltech) - disk block reading and tape writing
94 * are exported to several slave processes. While one slave writes the
95 * tape, the others read disk blocks; they pass control of the tape in
96 * a ring via signals. The parent process traverses the filesystem and
97 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
98 * The following structure defines the instruction packets sent to slaves.
106 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
108 int tapea; /* header number at start of this chunk */
109 int count; /* count to next header (used for TS_TAPE */
111 int inode; /* inode that we are currently dealing with */
112 int fd; /* FD for this slave */
113 int pid; /* PID for this slave */
114 int sent; /* 1 == we've sent this slave requests */
115 int firstrec; /* record number of this block */
116 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
117 struct req *req; /* buffer for requests */
121 char (*nextblock)[TP_BSIZE];
123 int master; /* pid of master, for sending error signals */
124 int tenths; /* length of tape used per block written */
125 static int caught; /* have we caught the signal to proceed? */
126 static int ready; /* have we reached the lock point without having */
127 /* received the SIGUSR2 signal from the prev slave? */
128 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */
129 /* SIGUSR2 arrives from the previous slave */
134 int pgoff = getpagesize() - 1;
138 writesize = ntrec * TP_BSIZE;
139 reqsiz = (ntrec + 1) * sizeof(struct req);
141 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
142 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
143 * repositioning after stopping, i.e, streaming mode, where the gap is
144 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
146 if (blocksperfile == 0 && !unlimited)
147 tenths = writesize / density +
148 (cartridge ? 16 : density == 625 ? 5 : 8);
150 * Allocate tape buffer contiguous with the array of instruction
151 * packets, so flushtape() can write them together with one write().
152 * Align tape buffer on page boundary to speed up tape write().
154 for (i = 0; i <= SLAVES; i++) {
156 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
159 slaves[i].tblock = (char (*)[TP_BSIZE])
160 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
161 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
167 nextblock = slp->tblock;
172 writerec(char *dp, int isspcl)
175 slp->req[trecno].dblk = (daddr_t)0;
176 slp->req[trecno].count = 1;
178 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp;
180 bcopy(dp, *(nextblock)++, sizeof (union u_spcl));
183 lastspclrec = spcl.c_tapea;
191 dumpblock(daddr_t blkno, int size)
193 int avail, tpblks, dblkno;
195 dblkno = fsbtodb(sblock, blkno);
196 tpblks = size >> tp_bshift;
197 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
198 slp->req[trecno].dblk = dblkno;
199 slp->req[trecno].count = avail;
201 spcl.c_tapea += avail;
204 dblkno += avail << (tp_bshift - dev_bshift);
212 tperror(int signo __unused)
216 msg("write error on %s\n", tape);
217 quit("Cannot recover\n");
220 msg("write error %ld 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");
233 sigpipe(int signo __unused)
236 quit("Broken pipe\n");
245 int siz = (char *)nextblock - (char *)slp->req;
247 slp->req[trecno].count = 0; /* Sentinel */
249 if (atomic_write(slp->fd, slp->req, siz) != siz)
250 quit("error writing command pipe: %s\n", strerror(errno));
251 slp->sent = 1; /* we sent a request, read the response later */
253 lastfirstrec = slp->firstrec;
255 if (++slp >= &slaves[SLAVES])
258 /* Read results back from next slave */
260 if (atomic_read(slp->fd, &got, sizeof got)
262 perror(" DUMP: error reading command pipe in master");
267 /* Check for end of tape */
268 if (got < writesize) {
269 msg("End of tape detected\n");
272 * Drain the results, don't care what the values were.
273 * If we read them here then trewind won't...
275 for (i = 0; i < SLAVES; i++) {
276 if (slaves[i].sent) {
277 if (atomic_read(slaves[i].fd,
280 perror(" DUMP: error reading command pipe in master");
294 if (spcl.c_type != TS_END) {
295 for (i = 0; i < spcl.c_count; i++)
296 if (spcl.c_addr[i] != 0)
299 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
300 slp->tapea = spcl.c_tapea;
301 slp->firstrec = lastfirstrec + ntrec;
303 nextblock = slp->tblock;
306 blockswritten += ntrec;
307 blocksthisvol += ntrec;
308 if (!pipeout && !unlimited && (blocksperfile ?
309 (blocksthisvol >= blocksperfile) : (asize > tsize))) {
323 for (f = 0; f < SLAVES; f++) {
325 * Drain the results, but unlike EOT we DO (or should) care
326 * what the return values were, since if we detect EOT after
327 * we think we've written the last blocks to the tape anyway,
328 * we have to replay those blocks with rollforward.
330 * fixme: punt for now.
332 if (slaves[f].sent) {
333 if (atomic_read(slaves[f].fd, &got, sizeof got)
335 perror(" DUMP: error reading command pipe in master");
339 if (got != writesize) {
340 msg("EOT detected in last 2 tape records!\n");
341 msg("Use a longer tape, decrease the size estimate\n");
342 quit("or use no size estimate at all.\n");
347 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
353 msg("Closing %s\n", tape);
358 while (rmtopen(tape, 0) < 0)
364 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) {
369 while ((f = open(tape, 0)) < 0)
377 time_t tstart_changevol, tend_changevol;
382 time((time_t *)&(tstart_changevol));
384 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
385 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
387 while (!query("Is the new volume mounted and ready to go?"))
388 if (query("Do you want to abort?")) {
392 time((time_t *)&(tend_changevol));
393 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1))
394 tstart_writing += (tend_changevol - tstart_changevol);
400 struct req *p, *q, *prev;
402 int i, size, savedtapea, got;
403 union u_spcl *ntb, *otb;
404 tslp = &slaves[SLAVES];
405 ntb = (union u_spcl *)tslp->tblock[1];
408 * Each of the N slaves should have requests that need to
409 * be replayed on the next tape. Use the extra slave buffers
410 * (slaves[SLAVES]) to construct request lists to be sent to
411 * each slave in turn.
413 for (i = 0; i < SLAVES; i++) {
415 otb = (union u_spcl *)slp->tblock;
418 * For each request in the current slave, copy it to tslp.
422 for (p = slp->req; p->count > 0; p += p->count) {
425 *ntb++ = *otb++; /* copy the datablock also */
430 quit("rollforward: protocol botch");
442 nextblock = tslp->tblock;
443 savedtapea = spcl.c_tapea;
444 spcl.c_tapea = slp->tapea;
446 spcl.c_tapea = savedtapea;
447 lastspclrec = savedtapea - 1;
449 size = (char *)ntb - (char *)q;
450 if (atomic_write(slp->fd, q, size) != size) {
451 perror(" DUMP: error writing command pipe");
455 if (++slp >= &slaves[SLAVES])
460 if (prev->dblk != 0) {
462 * If the last one was a disk block, make the
463 * first of this one be the last bit of that disk
466 q->dblk = prev->dblk +
467 prev->count * (TP_BSIZE / DEV_BSIZE);
468 ntb = (union u_spcl *)tslp->tblock;
471 * It wasn't a disk block. Copy the data to its
472 * new location in the buffer.
475 *((union u_spcl *)tslp->tblock) = *ntb;
476 ntb = (union u_spcl *)tslp->tblock[1];
480 nextblock = slp->tblock;
486 * Clear the first slaves' response. One hopes that it
487 * worked ok, otherwise the tape is much too short!
490 if (atomic_read(slp->fd, &got, sizeof got)
492 perror(" DUMP: error reading command pipe in master");
497 if (got != writesize) {
498 quit("EOT detected at start of the tape!\n");
504 * We implement taking and restoring checkpoints on the tape level.
505 * When each tape is opened, a new process is created by forking; this
506 * saves all of the necessary context in the parent. The child
507 * continues the dump; the parent waits around, saving the context.
508 * If the child returns X_REWRITE, then it had problems writing that tape;
509 * this causes the parent to fork again, duplicating the context, and
510 * everything continues as if nothing had happened.
513 startnewtape(int top)
521 void (*interrupt_save)();
523 sig_t interrupt_save;
526 interrupt_save = signal(SIGINT, SIG_IGN);
527 parentpid = getpid();
530 signal(SIGINT, interrupt_save);
532 * All signals are inherited...
534 setproctitle(NULL); /* Restore the proctitle. */
537 msg("Context save fork fails in parent %d\n", parentpid);
543 * save the context by waiting
544 * until the child doing all of the work returns.
545 * don't catch the interrupt
547 signal(SIGINT, SIG_IGN);
549 msg("Tape: %d; parent process: %d child process %d\n",
550 tapeno+1, parentpid, childpid);
552 while ((wait_pid = wait(&status)) != childpid)
553 msg("Parent %d waiting for child %d has another child %d return\n",
554 parentpid, childpid, wait_pid);
556 msg("Child %d returns LOB status %o\n",
557 childpid, status&0xFF);
559 status = (status >> 8) & 0xFF;
563 msg("Child %d finishes X_FINOK\n", childpid);
566 msg("Child %d finishes X_ABORT\n", childpid);
569 msg("Child %d finishes X_REWRITE\n", childpid);
572 msg("Child %d finishes unknown %d\n",
583 goto restore_check_point;
585 msg("Bad return code from dump: %d\n", status);
589 } else { /* we are the child; just continue */
591 sleep(4); /* allow time for parent's message to get out */
592 msg("Child on Tape %d has parent %d, my pid = %d\n",
593 tapeno+1, parentpid, getpid());
596 * If we have a name like "/dev/rmt0,/dev/rmt1",
597 * use the name before the comma first, and save
598 * the remaining names for subsequent volumes.
600 tapeno++; /* current tape sequence */
601 if (nexttape || strchr(tape, ',')) {
602 if (nexttape && *nexttape)
604 if ((p = strchr(tape, ',')) != NULL) {
609 msg("Dumping volume %d on %s\n", tapeno, tape);
612 while ((tapefd = (host ? rmtopen(tape, 2) :
613 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
615 while ((tapefd = (pipeout ? 1 :
616 open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
619 msg("Cannot open output \"%s\".\n", tape);
620 if (!query("Do you want to retry the open?"))
624 enslave(); /* Share open tape file descriptor with slaves */
625 signal(SIGINFO, infosch);
630 newtape++; /* new tape signal */
631 spcl.c_count = slp->count;
633 * measure firstrec in TP_BSIZE units since restore doesn't
634 * know the correct ntrec value...
636 spcl.c_firstrec = slp->firstrec;
638 spcl.c_type = TS_TAPE;
639 spcl.c_flags |= DR_NEWHEADER;
640 writeheader((ino_t)slp->inode);
641 spcl.c_flags &=~ DR_NEWHEADER;
643 msg("Volume %d begins with blocks from inode %d\n",
649 dumpabort(int signo __unused)
652 if (master != 0 && master != getpid())
653 /* Signals master to call dumpabort */
654 kill(master, SIGTERM);
657 msg("The ENTIRE dump is aborted.\n");
670 msg("pid = %d exits with status %d\n", getpid(), status);
676 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
679 proceed(int signo __unused)
695 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */
696 signal(SIGPIPE, sigpipe);
697 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */
698 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */
700 for (i = 0; i < SLAVES; i++) {
701 if (i == slp - &slaves[0]) {
707 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
708 (slaves[i].pid = fork()) < 0)
709 quit("too many slaves, %d (recompile smaller): %s\n",
712 slaves[i].fd = cmd[1];
714 if (slaves[i].pid == 0) { /* Slave starts up here */
715 for (j = 0; j <= i; j++)
717 signal(SIGINT, SIG_IGN); /* Master handles this */
727 for (i = 0; i < SLAVES; i++)
728 atomic_write(slaves[i].fd, &slaves[(i + 1) % SLAVES].pid,
729 sizeof slaves[0].pid);
739 for (i = 0; i < SLAVES; i++)
740 if (slaves[i].pid > 0) {
741 kill(slaves[i].pid, SIGKILL);
747 * Synchronization - each process has a lockfile, and shares file
748 * descriptors to the following process's lockfile. When our write
749 * completes, we release our lock on the following process's lock-
750 * file, allowing the following process to lock it and proceed. We
751 * get the lock back for the next cycle by swapping descriptors.
755 doslave(int cmd, int slave_number)
761 int nextslave, size, wrote, eot_count;
764 * Need our own seek pointer.
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, &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, slp->req, reqsiz)) == reqsiz) {
782 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);
791 atomic_read(cmd, 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... */
809 while (eot_count < 10 && size < writesize) {
812 wrote = rmtwrite(slp->tblock[0]+size,
816 wrote = write(tapefd, slp->tblock[0]+size,
819 printf("slave %d wrote %d\n", slave_number, wrote);
829 if (size != writesize)
830 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
831 slave_number, size, writesize);
835 * Handle ENOSPC as an EOT condition.
837 if (wrote < 0 && errno == ENOSPC) {
846 kill(master, SIGUSR1);
851 * pass size of write back to master
854 atomic_write(cmd, &size, sizeof size);
858 * If partial write, don't want next slave to go.
859 * Also jolts him awake.
861 kill(nextslave, SIGUSR2);
864 quit("error reading command pipe: %s\n", strerror(errno));
868 atomic_read(int fd, void *buf, int count)
870 int got, need = count;
872 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
873 buf = (uint8_t *)buf + got;
874 return (got < 0 ? got : count - need);
878 atomic_write(int fd, const void *buf, int count)
880 int got, need = count;
882 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
883 buf = (uint8_t *)buf + got;
884 return (got < 0 ? got : count - need);