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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tape.c 8.4 (Berkeley) 5/1/95
30 * $FreeBSD: src/sbin/dump/tape.c,v 1.12.2.3 2002/02/23 22:32:51 iedowse Exp $
33 #include <sys/param.h>
34 #include <sys/socket.h>
39 #include <sys/vnode.h>
42 #include <ufs/inode.h>
44 #include <vfs/ufs/dinode.h>
45 #include <vfs/ufs/fs.h>
48 #include <protocols/dumprestore.h>
61 static int writesize; /* size of malloc()ed buffer for tape */
62 static long lastspclrec = -1; /* tape block number of last written header */
63 static int trecno = 0; /* next record to write in current block */
64 static long blocksthisvol; /* number of blocks on current output file */
65 static const char *nexttape;
66 static int tapeno = 0; /* current tape number */
68 static int atomic_read(int, void *, int);
69 static int atomic_write(int, const void *, int);
71 static void doslave(int, int);
73 static void doslave(int);
75 static void enslave(void);
76 static void flushtape(void);
77 static void killall(void);
78 static void rollforward(void);
81 * Concurrent dump mods (Caltech) - disk block reading and tape writing
82 * are exported to several slave processes. While one slave writes the
83 * tape, the others read disk blocks; they pass control of the tape in
84 * a ring via signals. The parent process traverses the filesystem and
85 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
86 * The following structure defines the instruction packets sent to slaves.
94 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
96 int tapea; /* header number at start of this chunk */
97 int count; /* count to next header (used for TS_TAPE */
99 int inode; /* inode that we are currently dealing with */
100 int fd; /* FD for this slave */
101 int pid; /* PID for this slave */
102 int sent; /* 1 == we've sent this slave requests */
103 int firstrec; /* record number of this block */
104 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
105 struct req *req; /* buffer for requests */
109 char (*nextblock)[TP_BSIZE];
111 int master; /* pid of master, for sending error signals */
112 int tenths; /* length of tape used per block written */
113 static int caught; /* have we caught the signal to proceed? */
114 static int ready; /* have we reached the lock point without having */
115 /* received the SIGUSR2 signal from the prev slave? */
116 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */
117 /* SIGUSR2 arrives from the previous slave */
122 int pgoff = getpagesize() - 1;
126 writesize = ntrec * TP_BSIZE;
127 reqsiz = (ntrec + 1) * sizeof(struct req);
129 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
130 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
131 * repositioning after stopping, i.e, streaming mode, where the gap is
132 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
134 if (blocksperfile == 0 && !unlimited)
135 tenths = writesize / density +
136 (cartridge ? 16 : density == 625 ? 5 : 8);
138 * Allocate tape buffer contiguous with the array of instruction
139 * packets, so flushtape() can write them together with one write().
140 * Align tape buffer on page boundary to speed up tape write().
142 for (i = 0; i <= SLAVES; i++) {
144 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
147 slaves[i].tblock = (char (*)[TP_BSIZE])
148 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
149 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
155 nextblock = slp->tblock;
160 writerec(const void *dp, int isspcl)
163 slp->req[trecno].dblk = (daddr_t)0;
164 slp->req[trecno].count = 1;
165 bcopy(dp, *(nextblock)++, sizeof (union u_spcl));
167 lastspclrec = spcl.c_tapea;
175 dumpblock(daddr_t blkno, int size)
177 int avail, tpblks, dblkno;
179 dblkno = fsbtodb(sblock, blkno);
180 tpblks = size >> tp_bshift;
181 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
182 slp->req[trecno].dblk = dblkno;
183 slp->req[trecno].count = avail;
185 spcl.c_tapea += avail;
188 dblkno += avail << (tp_bshift - dev_bshift);
196 tperror(int signo __unused)
200 msg("write error on %s\n", tape);
201 quit("Cannot recover\n");
204 msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno);
205 broadcast("DUMP WRITE ERROR!\n");
206 if (!query("Do you want to restart?"))
208 msg("Closing this volume. Prepare to restart with new media;\n");
209 msg("this dump volume will be rewritten.\n");
217 sigpipe(int signo __unused)
220 quit("Broken pipe\n");
229 int siz = (char *)nextblock - (char *)slp->req;
231 slp->req[trecno].count = 0; /* Sentinel */
233 if (atomic_write(slp->fd, slp->req, siz) != siz)
234 quit("error writing command pipe: %s\n", strerror(errno));
235 slp->sent = 1; /* we sent a request, read the response later */
237 lastfirstrec = slp->firstrec;
239 if (++slp >= &slaves[SLAVES])
242 /* Read results back from next slave */
244 if (atomic_read(slp->fd, &got, sizeof got)
246 perror(" DUMP: error reading command pipe in master");
251 /* Check for end of tape */
252 if (got < writesize) {
253 msg("End of tape detected\n");
256 * Drain the results, don't care what the values were.
257 * If we read them here then trewind won't...
259 for (i = 0; i < SLAVES; i++) {
260 if (slaves[i].sent) {
261 if (atomic_read(slaves[i].fd,
264 perror(" DUMP: error reading command pipe in master");
278 if (spcl.c_type != TS_END) {
279 for (i = 0; i < spcl.c_count; i++)
280 if (spcl.c_addr[i] != 0)
283 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
284 slp->tapea = spcl.c_tapea;
285 slp->firstrec = lastfirstrec + ntrec;
287 nextblock = slp->tblock;
290 blockswritten += ntrec;
291 blocksthisvol += ntrec;
292 if (!pipeout && !unlimited && (blocksperfile ?
293 (blocksthisvol >= blocksperfile) : (asize > tsize))) {
307 for (f = 0; f < SLAVES; f++) {
309 * Drain the results, but unlike EOT we DO (or should) care
310 * what the return values were, since if we detect EOT after
311 * we think we've written the last blocks to the tape anyway,
312 * we have to replay those blocks with rollforward.
314 * fixme: punt for now.
316 if (slaves[f].sent) {
317 if (atomic_read(slaves[f].fd, &got, sizeof got)
319 perror(" DUMP: error reading command pipe in master");
323 if (got != writesize) {
324 msg("EOT detected in last 2 tape records!\n");
325 msg("Use a longer tape, decrease the size estimate\n");
326 quit("or use no size estimate at all.\n");
331 while (wait(NULL) >= 0) /* wait for any signals from slaves */
337 msg("Closing %s\n", tape);
342 while (rmtopen(tape, 0) < 0)
348 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) {
353 while ((f = open(tape, 0)) < 0)
361 time_t tstart_changevol, tend_changevol;
366 time((time_t *)&(tstart_changevol));
368 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
369 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
371 while (!query("Is the new volume mounted and ready to go?"))
372 if (query("Do you want to abort?")) {
376 time((time_t *)&(tend_changevol));
377 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1))
378 tstart_writing += (tend_changevol - tstart_changevol);
384 struct req *p, *q, *prev;
386 int i, size, savedtapea, got;
387 union u_spcl *ntb, *otb;
388 tslp = &slaves[SLAVES];
389 ntb = (union u_spcl *)tslp->tblock[1];
392 * Each of the N slaves should have requests that need to
393 * be replayed on the next tape. Use the extra slave buffers
394 * (slaves[SLAVES]) to construct request lists to be sent to
395 * each slave in turn.
397 for (i = 0; i < SLAVES; i++) {
399 otb = (union u_spcl *)slp->tblock;
402 * For each request in the current slave, copy it to tslp.
406 for (p = slp->req; p->count > 0; p += p->count) {
409 *ntb++ = *otb++; /* copy the datablock also */
414 quit("rollforward: protocol botch");
426 nextblock = tslp->tblock;
427 savedtapea = spcl.c_tapea;
428 spcl.c_tapea = slp->tapea;
430 spcl.c_tapea = savedtapea;
431 lastspclrec = savedtapea - 1;
433 size = (char *)ntb - (char *)q;
434 if (atomic_write(slp->fd, q, size) != size) {
435 perror(" DUMP: error writing command pipe");
439 if (++slp >= &slaves[SLAVES])
444 if (prev->dblk != 0) {
446 * If the last one was a disk block, make the
447 * first of this one be the last bit of that disk
450 q->dblk = prev->dblk +
451 prev->count * (TP_BSIZE / DEV_BSIZE);
452 ntb = (union u_spcl *)tslp->tblock;
455 * It wasn't a disk block. Copy the data to its
456 * new location in the buffer.
459 *((union u_spcl *)tslp->tblock) = *ntb;
460 ntb = (union u_spcl *)tslp->tblock[1];
464 nextblock = slp->tblock;
470 * Clear the first slaves' response. One hopes that it
471 * worked ok, otherwise the tape is much too short!
474 if (atomic_read(slp->fd, &got, sizeof got)
476 perror(" DUMP: error reading command pipe in master");
481 if (got != writesize) {
482 quit("EOT detected at start of the tape!\n");
488 * We implement taking and restoring checkpoints on the tape level.
489 * When each tape is opened, a new process is created by forking; this
490 * saves all of the necessary context in the parent. The child
491 * continues the dump; the parent waits around, saving the context.
492 * If the child returns X_REWRITE, then it had problems writing that tape;
493 * this causes the parent to fork again, duplicating the context, and
494 * everything continues as if nothing had happened.
497 startnewtape(int top)
505 void (*interrupt_save)();
507 sig_t interrupt_save;
510 interrupt_save = signal(SIGINT, SIG_IGN);
511 parentpid = getpid();
514 signal(SIGINT, interrupt_save);
516 * All signals are inherited...
518 setproctitle(NULL); /* Restore the proctitle. */
521 msg("Context save fork fails in parent %d\n", parentpid);
527 * save the context by waiting
528 * until the child doing all of the work returns.
529 * don't catch the interrupt
531 signal(SIGINT, SIG_IGN);
533 msg("Tape: %d; parent process: %d child process %d\n",
534 tapeno+1, parentpid, childpid);
536 while ((wait_pid = wait(&status)) != childpid)
537 msg("Parent %d waiting for child %d has another child %d return\n",
538 parentpid, childpid, wait_pid);
540 msg("Child %d returns LOB status %o\n",
541 childpid, status&0xFF);
543 status = (status >> 8) & 0xFF;
547 msg("Child %d finishes X_FINOK\n", childpid);
550 msg("Child %d finishes X_ABORT\n", childpid);
553 msg("Child %d finishes X_REWRITE\n", childpid);
556 msg("Child %d finishes unknown %d\n",
567 goto restore_check_point;
569 msg("Bad return code from dump: %d\n", status);
573 } else { /* we are the child; just continue */
575 sleep(4); /* allow time for parent's message to get out */
576 msg("Child on Tape %d has parent %d, my pid = %d\n",
577 tapeno+1, parentpid, getpid());
580 * If we have a name like "/dev/rmt0,/dev/rmt1",
581 * use the name before the comma first, and save
582 * the remaining names for subsequent volumes.
584 tapeno++; /* current tape sequence */
585 if (nexttape || strchr(tape, ',')) {
586 if (nexttape && *nexttape)
588 if ((p = strchr(tape, ',')) != NULL) {
593 msg("Dumping volume %d on %s\n", tapeno, tape);
596 while ((tapefd = (host ? rmtopen(tape, 2) :
597 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
599 while ((tapefd = (pipeout ? 1 :
600 open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
603 msg("Cannot open output \"%s\".\n", tape);
604 if (!query("Do you want to retry the open?"))
608 enslave(); /* Share open tape file descriptor with slaves */
609 signal(SIGINFO, infosch);
614 newtape++; /* new tape signal */
615 spcl.c_count = slp->count;
617 * measure firstrec in TP_BSIZE units since restore doesn't
618 * know the correct ntrec value...
620 spcl.c_firstrec = slp->firstrec;
622 spcl.c_type = TS_TAPE;
623 spcl.c_flags |= DR_NEWHEADER;
624 writeheader((ufs1_ino_t)slp->inode);
625 spcl.c_flags &=~ DR_NEWHEADER;
627 msg("Volume %d begins with blocks from inode %d\n",
633 dumpabort(int signo __unused)
636 if (master != 0 && master != getpid())
637 /* Signals master to call dumpabort */
638 kill(master, SIGTERM);
641 msg("The ENTIRE dump is aborted.\n");
654 msg("pid = %d exits with status %d\n", getpid(), status);
660 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
663 proceed(int signo __unused)
679 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */
680 signal(SIGPIPE, sigpipe);
681 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */
682 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */
684 for (i = 0; i < SLAVES; i++) {
685 if (i == slp - &slaves[0]) {
691 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
692 (slaves[i].pid = fork()) < 0)
693 quit("too many slaves, %d (recompile smaller): %s\n",
696 slaves[i].fd = cmd[1];
698 if (slaves[i].pid == 0) { /* Slave starts up here */
699 for (j = 0; j <= i; j++)
701 signal(SIGINT, SIG_IGN); /* Master handles this */
711 for (i = 0; i < SLAVES; i++)
712 atomic_write(slaves[i].fd, &slaves[(i + 1) % SLAVES].pid,
713 sizeof slaves[0].pid);
723 for (i = 0; i < SLAVES; i++)
724 if (slaves[i].pid > 0) {
725 kill(slaves[i].pid, SIGKILL);
731 * Synchronization - each process has a lockfile, and shares file
732 * descriptors to the following process's lockfile. When our write
733 * completes, we release our lock on the following process's lock-
734 * file, allowing the following process to lock it and proceed. We
735 * get the lock back for the next cycle by swapping descriptors.
739 doslave(int cmd, int slave_number)
745 int nextslave, size, wrote, eot_count;
748 * Need our own seek pointer.
751 if ((diskfd = open(disk, O_RDONLY)) < 0)
752 quit("slave couldn't reopen disk: %s\n", strerror(errno));
755 * Need the pid of the next slave in the loop...
757 if ((nread = atomic_read(cmd, &nextslave, sizeof nextslave))
758 != sizeof nextslave) {
759 quit("master/slave protocol botched - didn't get pid of next slave.\n");
763 * Get list of blocks to dump, read the blocks into tape buffer
765 while ((nread = atomic_read(cmd, slp->req, reqsiz)) == reqsiz) {
766 struct req *p = slp->req;
768 for (trecno = 0; trecno < ntrec;
769 trecno += p->count, p += p->count) {
771 bread(p->dblk, slp->tblock[trecno],
772 p->count * TP_BSIZE);
775 atomic_read(cmd, slp->tblock[trecno],
776 TP_BSIZE) != TP_BSIZE)
777 quit("master/slave protocol botched.\n");
780 if (setjmp(jmpbuf) == 0) {
788 /* Try to write the data... */
793 while (eot_count < 10 && size < writesize) {
796 wrote = rmtwrite(slp->tblock[0]+size,
800 wrote = write(tapefd, slp->tblock[0]+size,
803 printf("slave %d wrote %d\n", slave_number, wrote);
813 if (size != writesize)
814 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
815 slave_number, size, writesize);
819 * Handle ENOSPC as an EOT condition.
821 if (wrote < 0 && errno == ENOSPC) {
830 kill(master, SIGUSR1);
835 * pass size of write back to master
838 atomic_write(cmd, &size, sizeof size);
842 * If partial write, don't want next slave to go.
843 * Also jolts him awake.
845 kill(nextslave, SIGUSR2);
848 quit("error reading command pipe: %s\n", strerror(errno));
852 atomic_read(int fd, void *buf, int count)
854 int got, need = count;
856 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
857 buf = (uint8_t *)buf + got;
858 return (got < 0 ? got : count - need);
862 atomic_write(int fd, const void *buf, int count)
864 int got, need = count;
866 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
867 buf = (const uint8_t *)buf + got;
868 return (got < 0 ? got : count - need);