1/*	$NetBSD: tape.c,v 1.49 2008/02/16 17:58:01 matt Exp $	*/
2
3/*-
4 * Copyright (c) 1980, 1991, 1993
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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.
18 *
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
29 * SUCH DAMAGE.
30 */
31
32#include <sys/cdefs.h>
33#ifndef lint
34#if 0
35static char sccsid[] = "@(#)tape.c	8.4 (Berkeley) 5/1/95";
36#else
37__RCSID("$NetBSD: tape.c,v 1.49 2008/02/16 17:58:01 matt Exp $");
38#endif
39#endif /* not lint */
40
41#include <sys/param.h>
42#include <sys/socket.h>
43#include <sys/time.h>
44#include <sys/wait.h>
45#include <ufs/ufs/dinode.h>
46#include <sys/ioctl.h>
47#include <sys/mtio.h>
48
49#include <protocols/dumprestore.h>
50
51#include <errno.h>
52#include <fcntl.h>
53#include <signal.h>
54#include <stdio.h>
55#include <stdlib.h>
56#include <string.h>
57#include <time.h>
58#include <unistd.h>
59
60#include "dump.h"
61#include "pathnames.h"
62
63int	writesize;		/* size of malloc()ed buffer for tape */
64int64_t	lastspclrec = -1;	/* tape block number of last written header */
65int	trecno = 0;		/* next record to write in current block */
66extern	long blocksperfile;	/* number of blocks per output file */
67long	blocksthisvol;		/* number of blocks on current output file */
68extern	int ntrec;		/* blocking factor on tape */
69extern	int cartridge;
70extern	const char *host;
71char	*nexttape;
72
73static	ssize_t atomic_read(int, char *, int);
74static	ssize_t atomic_write(int, char *, int);
75static	void doslave(int, int);
76static	void enslave(void);
77static	void flushtape(void);
78static	void killall(void);
79static	void proceed(int);
80static	void rollforward(void);
81static	void sigpipe(int);
82static	void tperror(int);
83
84/*
85 * Concurrent dump mods (Caltech) - disk block reading and tape writing
86 * are exported to several slave processes.  While one slave writes the
87 * tape, the others read disk blocks; they pass control of the tape in
88 * a ring via signals. The parent process traverses the file system and
89 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
90 * The following structure defines the instruction packets sent to slaves.
91 */
92struct req {
93	daddr_t dblk;
94	int count;
95};
96int reqsiz;
97
98#define SLAVES 3		/* 1 slave writing, 1 reading, 1 for slack */
99struct slave {
100	int64_t tapea;		/* header number at start of this chunk */
101	int64_t firstrec;	/* record number of this block */
102	int count;		/* count to next header (used for TS_TAPE */
103				/* after EOT) */
104	int inode;		/* inode that we are currently dealing with */
105	int fd;			/* FD for this slave */
106	int pid;		/* PID for this slave */
107	int sent;		/* 1 == we've sent this slave requests */
108	char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
109	struct req *req;	/* buffer for requests */
110} slaves[SLAVES+1];
111struct slave *slp;
112
113char	(*nextblock)[TP_BSIZE];
114
115static int64_t tapea_volume;	/* value of spcl.c_tapea at volume start */
116
117int master;		/* pid of master, for sending error signals */
118int tenths;		/* length of tape used per block written */
119static volatile sig_atomic_t caught;	/* have we caught the signal to proceed? */
120
121int
122alloctape(void)
123{
124	int pgoff = getpagesize() - 1;
125	char *buf;
126	int i;
127
128	writesize = ntrec * TP_BSIZE;
129	reqsiz = (ntrec + 1) * sizeof(struct req);
130	/*
131	 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
132	 * (see DEC TU80 User's Guide).  The shorter gaps of 6250-bpi require
133	 * repositioning after stopping, i.e, streaming mode, where the gap is
134	 * variable, 0.30" to 0.45".  The gap is maximal when the tape stops.
135	 */
136	if (blocksperfile == 0 && !unlimited)
137		tenths = writesize / density +
138		    (cartridge ? 16 : density == 625 ? 5 : 8);
139	/*
140	 * Allocate tape buffer contiguous with the array of instruction
141	 * packets, so flushtape() can write them together with one write().
142	 * Align tape buffer on page boundary to speed up tape write().
143	 */
144	for (i = 0; i <= SLAVES; i++) {
145		buf = (char *)
146		    xmalloc((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;
150	}
151	slp = &slaves[0];
152	slp->count = 1;
153	slp->tapea = 0;
154	slp->firstrec = 0;
155	nextblock = slp->tblock;
156	return(1);
157}
158
159void
160writerec(char *dp, int isspcl)
161{
162
163	slp->req[trecno].dblk = (daddr_t)0;
164	slp->req[trecno].count = 1;
165	*(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp;
166	if (isspcl)
167		lastspclrec = iswap64(spcl.c_tapea);
168	trecno++;
169	spcl.c_tapea = iswap64(iswap64(spcl.c_tapea) +1);
170	if (trecno >= ntrec)
171		flushtape();
172}
173
174void
175dumpblock(daddr_t blkno, int size)
176{
177	int avail, tpblks;
178	daddr_t dblkno;
179
180	dblkno = fsatoda(ufsib, blkno);
181	tpblks = size >> tp_bshift;
182	while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
183		slp->req[trecno].dblk = dblkno;
184		slp->req[trecno].count = avail;
185		trecno += avail;
186		spcl.c_tapea = iswap64(iswap64(spcl.c_tapea) + avail);
187		if (trecno >= ntrec)
188			flushtape();
189		dblkno += avail << (tp_bshift - dev_bshift);
190		tpblks -= avail;
191	}
192}
193
194int	nogripe = 0;
195
196static void
197tperror(int signo __unused)
198{
199
200	if (pipeout) {
201		msg("write error on %s\n", tape);
202		quit("Cannot recover\n");
203		/* NOTREACHED */
204	}
205	msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno);
206	broadcast("DUMP WRITE ERROR!\n");
207	if (!query("Do you want to restart?"))
208		dumpabort(0);
209	msg("Closing this volume.  Prepare to restart with new media;\n");
210	msg("this dump volume will be rewritten.\n");
211	killall();
212	nogripe = 1;
213	close_rewind();
214	Exit(X_REWRITE);
215}
216
217static void
218sigpipe(int signo __unused)
219{
220
221	quit("Broken pipe\n");
222}
223
224/*
225 * do_stats --
226 *	Update xferrate stats
227 */
228time_t
229do_stats(void)
230{
231	time_t tnow, ttaken;
232	int64_t blocks;
233
234	(void)time(&tnow);
235	ttaken = tnow - tstart_volume;
236	blocks = iswap64(spcl.c_tapea) - tapea_volume;
237	msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
238	if (ttaken > 0) {
239		msg("Volume %d took %d:%02d:%02d\n", tapeno,
240		    (int) (ttaken / 3600), (int) ((ttaken % 3600) / 60),
241		    (int) (ttaken % 60));
242		msg("Volume %d transfer rate: %d KB/s\n", tapeno,
243		    (int) (blocks / ttaken));
244		xferrate += blocks / ttaken;
245	}
246	return(tnow);
247}
248
249/*
250 * statussig --
251 *	information message upon receipt of SIGINFO
252 *	(derived from optr.c::timeest())
253 */
254void
255statussig(int notused __unused)
256{
257	time_t	tnow, deltat;
258	char	msgbuf[128];
259	int	errno_save;
260
261	if (blockswritten < 500)
262		return;
263	errno_save = errno;
264	(void) time((time_t *) &tnow);
265	if (tnow <= tstart_volume)
266		return;
267	deltat = tstart_writing - tnow +
268	    (1.0 * (tnow - tstart_writing)) / blockswritten * tapesize;
269	(void)snprintf(msgbuf, sizeof(msgbuf),
270	    "%3.2f%% done at %ld KB/s, finished in %d:%02d\n",
271	    (blockswritten * 100.0) / tapesize,
272	    (long)((iswap64(spcl.c_tapea) - tapea_volume) /
273	      (tnow - tstart_volume)),
274	    (int)(deltat / 3600), (int)((deltat % 3600) / 60));
275	write(STDERR_FILENO, msgbuf, strlen(msgbuf));
276	errno = errno_save;
277}
278
279static void
280flushtape(void)
281{
282	int i, blks, got;
283	int64_t lastfirstrec;
284
285	int siz = (char *)nextblock - (char *)slp->req;
286
287	slp->req[trecno].count = 0;			/* Sentinel */
288
289	if (atomic_write(slp->fd, (char *)slp->req, siz) != siz)
290		quit("error writing command pipe: %s\n", strerror(errno));
291	slp->sent = 1; /* we sent a request, read the response later */
292
293	lastfirstrec = slp->firstrec;
294
295	if (++slp >= &slaves[SLAVES])
296		slp = &slaves[0];
297
298	/* Read results back from next slave */
299	if (slp->sent) {
300		if (atomic_read(slp->fd, (char *)&got, sizeof got)
301		    != sizeof got) {
302			perror("  DUMP: error reading command pipe in master");
303			dumpabort(0);
304		}
305		slp->sent = 0;
306
307		/* Check for end of tape */
308		if (got < writesize) {
309			msg("End of tape detected\n");
310
311			/*
312			 * Drain the results, don't care what the values were.
313			 * If we read them here then trewind won't...
314			 */
315			for (i = 0; i < SLAVES; i++) {
316				if (slaves[i].sent) {
317					if (atomic_read(slaves[i].fd,
318					    (char *)&got, sizeof got)
319					    != sizeof got) {
320						perror("  DUMP: error reading command pipe in master");
321						dumpabort(0);
322					}
323					slaves[i].sent = 0;
324				}
325			}
326
327			close_rewind();
328			rollforward();
329			return;
330		}
331	}
332
333	blks = 0;
334	if (iswap32(spcl.c_type) != TS_END) {
335		for (i = 0; i < iswap32(spcl.c_count); i++)
336			if (spcl.c_addr[i] != 0)
337				blks++;
338	}
339	slp->count = lastspclrec + blks + 1 - iswap32(spcl.c_tapea);
340	slp->tapea = iswap64(spcl.c_tapea);
341	slp->firstrec = lastfirstrec + ntrec;
342	slp->inode = curino;
343	nextblock = slp->tblock;
344	trecno = 0;
345	asize += tenths;
346	blockswritten += ntrec;
347	blocksthisvol += ntrec;
348	if (!pipeout && !unlimited && (blocksperfile ?
349	    (blocksthisvol >= blocksperfile) : (asize > tsize))) {
350		close_rewind();
351		startnewtape(0);
352	}
353	timeest();
354}
355
356void
357trewind(int eject)
358{
359	int f;
360	int got;
361
362	for (f = 0; f < SLAVES; f++) {
363		/*
364		 * Drain the results, but unlike EOT we DO (or should) care
365		 * what the return values were, since if we detect EOT after
366		 * we think we've written the last blocks to the tape anyway,
367		 * we have to replay those blocks with rollforward.
368		 *
369		 * fixme: punt for now.
370		 */
371		if (slaves[f].sent) {
372			if (atomic_read(slaves[f].fd, (char *)&got, sizeof got)
373			    != sizeof got) {
374				perror("  DUMP: error reading command pipe in master");
375				dumpabort(0);
376			}
377			slaves[f].sent = 0;
378			if (got != writesize) {
379				msg("EOT detected in last 2 tape records!\n");
380				msg("Use a longer tape, decrease the size estimate\n");
381				quit("or use no size estimate at all.\n");
382			}
383		}
384		(void) close(slaves[f].fd);
385	}
386	while (wait(NULL) >= 0)	/* wait for any signals from slaves */
387		/* void */;
388
389	if (pipeout)
390		return;
391
392	msg("Closing %s\n", tape);
393
394#ifdef RDUMP
395	if (host) {
396		rmtclose();
397		while (rmtopen(tape, 0, 0) < 0)
398			sleep(10);
399		if (eflag && eject) {
400			msg("Ejecting %s\n", tape);
401			(void) rmtioctl(MTOFFL, 0);
402		}
403		rmtclose();
404		return;
405	}
406#endif
407	(void) close(tapefd);
408	while ((f = open(tape, 0)) < 0)
409		sleep (10);
410	if (eflag && eject) {
411		struct mtop offl;
412
413		msg("Ejecting %s\n", tape);
414		offl.mt_op = MTOFFL;
415		offl.mt_count = 0;
416		(void) ioctl(f, MTIOCTOP, &offl);
417	}
418	(void) close(f);
419}
420
421void
422close_rewind(void)
423{
424	int i, f;
425
426	trewind(1);
427	(void)do_stats();
428	if (nexttape)
429		return;
430	if (!nogripe) {
431		msg("Change Volumes: Mount volume #%d\n", tapeno+1);
432		broadcast("CHANGE DUMP VOLUMES!\a\a\n");
433	}
434	if (lflag) {
435		for (i = 0; i < lflag / 10; i++) { /* wait lflag seconds */
436			if (host) {
437				if (rmtopen(tape, 0, 0) >= 0) {
438					rmtclose();
439					return;
440				}
441			} else {
442				if ((f = open(tape, 0)) >= 0) {
443					close(f);
444					return;
445				}
446			}
447			sleep (10);
448		}
449	}
450
451	while (!query("Is the new volume mounted and ready to go?"))
452		if (query("Do you want to abort?")) {
453			dumpabort(0);
454			/*NOTREACHED*/
455		}
456}
457
458void
459rollforward(void)
460{
461	struct req *p, *q, *prev;
462	struct slave *tslp;
463	int i, size, savedtapea, got;
464	union u_spcl *ntb, *otb;
465	tslp = &slaves[SLAVES];
466	ntb = (union u_spcl *)tslp->tblock[1];
467
468	/*
469	 * Each of the N slaves should have requests that need to
470	 * be replayed on the next tape.  Use the extra slave buffers
471	 * (slaves[SLAVES]) to construct request lists to be sent to
472	 * each slave in turn.
473	 */
474	for (i = 0; i < SLAVES; i++) {
475		q = &tslp->req[1];
476		otb = (union u_spcl *)slp->tblock;
477
478		/*
479		 * For each request in the current slave, copy it to tslp.
480		 */
481
482		prev = NULL;
483		for (p = slp->req; p->count > 0; p += p->count) {
484			*q = *p;
485			if (p->dblk == 0)
486				*ntb++ = *otb++; /* copy the datablock also */
487			prev = q;
488			q += q->count;
489		}
490		if (prev == NULL)
491			quit("rollforward: protocol botch");
492		if (prev->dblk != 0)
493			prev->count -= 1;
494		else
495			ntb--;
496		q -= 1;
497		q->count = 0;
498		q = &tslp->req[0];
499		if (i == 0) {
500			q->dblk = 0;
501			q->count = 1;
502			trecno = 0;
503			nextblock = tslp->tblock;
504			savedtapea = iswap32(spcl.c_tapea);
505			spcl.c_tapea = iswap32(slp->tapea);
506			startnewtape(0);
507			spcl.c_tapea = iswap32(savedtapea);
508			lastspclrec = savedtapea - 1;
509		}
510		size = (char *)ntb - (char *)q;
511		if (atomic_write(slp->fd, (char *)q, size) != size) {
512			perror("  DUMP: error writing command pipe");
513			dumpabort(0);
514		}
515		slp->sent = 1;
516		if (++slp >= &slaves[SLAVES])
517			slp = &slaves[0];
518
519		q->count = 1;
520
521		if (prev->dblk != 0) {
522			/*
523			 * If the last one was a disk block, make the
524			 * first of this one be the last bit of that disk
525			 * block...
526			 */
527			q->dblk = prev->dblk +
528				prev->count * (TP_BSIZE / DEV_BSIZE);
529			ntb = (union u_spcl *)tslp->tblock;
530		} else {
531			/*
532			 * It wasn't a disk block.  Copy the data to its
533			 * new location in the buffer.
534			 */
535			q->dblk = 0;
536			*((union u_spcl *)tslp->tblock) = *ntb;
537			ntb = (union u_spcl *)tslp->tblock[1];
538		}
539	}
540	slp->req[0] = *q;
541	nextblock = slp->tblock;
542	if (q->dblk == 0)
543		nextblock++;
544	trecno = 1;
545
546	/*
547	 * Clear the first slaves' response.  One hopes that it
548	 * worked ok, otherwise the tape is much too short!
549	 */
550	if (slp->sent) {
551		if (atomic_read(slp->fd, (char *)&got, sizeof got)
552		    != sizeof got) {
553			perror("  DUMP: error reading command pipe in master");
554			dumpabort(0);
555		}
556		slp->sent = 0;
557
558		if (got != writesize) {
559			quit("EOT detected at start of the tape!\n");
560		}
561	}
562}
563
564/*
565 * We implement taking and restoring checkpoints on the tape level.
566 * When each tape is opened, a new process is created by forking; this
567 * saves all of the necessary context in the parent.  The child
568 * continues the dump; the parent waits around, saving the context.
569 * If the child returns X_REWRITE, then it had problems writing that tape;
570 * this causes the parent to fork again, duplicating the context, and
571 * everything continues as if nothing had happened.
572 */
573void
574startnewtape(int top)
575{
576	int	parentpid;
577	int	childpid;
578	int	status;
579	int	waitforpid;
580	char	*p;
581	sig_t	interrupt_save;
582
583	interrupt_save = signal(SIGINT, SIG_IGN);
584	parentpid = getpid();
585	tapea_volume = iswap32(spcl.c_tapea);
586	(void)time(&tstart_volume);
587
588restore_check_point:
589	(void)signal(SIGINT, interrupt_save);
590	/*
591	 *	All signals are inherited...
592	 */
593	childpid = fork();
594	if (childpid < 0) {
595		msg("Context save fork fails in parent %d\n", parentpid);
596		Exit(X_ABORT);
597	}
598	if (childpid != 0) {
599		/*
600		 *	PARENT:
601		 *	save the context by waiting
602		 *	until the child doing all of the work returns.
603		 *	don't catch the interrupt
604		 */
605		signal(SIGINT, SIG_IGN);
606		signal(SIGINFO, SIG_IGN);	/* only want child's stats */
607#ifdef TDEBUG
608		msg("Tape: %d; parent process: %d child process %d\n",
609			tapeno+1, parentpid, childpid);
610#endif /* TDEBUG */
611		while ((waitforpid = wait(&status)) != childpid)
612			msg("Parent %d waiting for child %d has another child %d return\n",
613				parentpid, childpid, waitforpid);
614		if (status & 0xFF) {
615			msg("Child %d returns LOB status %o\n",
616				childpid, status&0xFF);
617		}
618		status = (status >> 8) & 0xFF;
619#ifdef TDEBUG
620		switch(status) {
621			case X_FINOK:
622				msg("Child %d finishes X_FINOK\n", childpid);
623				break;
624			case X_ABORT:
625				msg("Child %d finishes X_ABORT\n", childpid);
626				break;
627			case X_REWRITE:
628				msg("Child %d finishes X_REWRITE\n", childpid);
629				break;
630			default:
631				msg("Child %d finishes unknown %d\n",
632					childpid, status);
633				break;
634		}
635#endif /* TDEBUG */
636		switch(status) {
637			case X_FINOK:
638				Exit(X_FINOK);
639			case X_ABORT:
640				Exit(X_ABORT);
641			case X_REWRITE:
642				goto restore_check_point;
643			default:
644				msg("Bad return code from dump: %d\n", status);
645				Exit(X_ABORT);
646		}
647		/*NOTREACHED*/
648	} else {	/* we are the child; just continue */
649		signal(SIGINFO, statussig);	/* now want child's stats */
650#ifdef TDEBUG
651		sleep(4);	/* allow time for parent's message to get out */
652		msg("Child on Tape %d has parent %d, my pid = %d\n",
653			tapeno+1, parentpid, getpid());
654#endif /* TDEBUG */
655		/*
656		 * If we have a name like "/dev/rst0,/dev/rst1",
657		 * use the name before the comma first, and save
658		 * the remaining names for subsequent volumes.
659		 */
660		tapeno++;			/* current tape sequence */
661		if (nexttape || strchr(tape, ',')) {
662			if (nexttape && *nexttape)
663				tape = nexttape;
664			if ((p = strchr(tape, ',')) != NULL) {
665				*p = '\0';
666				nexttape = p + 1;
667			} else
668				nexttape = NULL;
669			msg("Dumping volume %d on %s\n", tapeno, tape);
670		}
671#ifdef RDUMP
672		while ((tapefd = (host ? rmtopen(tape, 2, 1) :
673			pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
674#else
675		while ((tapefd = (pipeout ? 1 :
676				  open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
677#endif
678		    {
679			msg("Cannot open output \"%s\".\n", tape);
680			if (!query("Do you want to retry the open?"))
681				dumpabort(0);
682		}
683
684		enslave();  /* Share open tape file descriptor with slaves */
685
686		asize = 0;
687		blocksthisvol = 0;
688		if (top)
689			newtape++;		/* new tape signal */
690		spcl.c_count = iswap32(slp->count);
691		/*
692		 * measure firstrec in TP_BSIZE units since restore doesn't
693		 * know the correct ntrec value...
694		 */
695		spcl.c_firstrec = iswap32(slp->firstrec);
696		spcl.c_volume = iswap32(iswap32(spcl.c_volume) + 1);
697		spcl.c_type = iswap32(TS_TAPE);
698		if (!is_ufs2)
699			spcl.c_flags = iswap32(iswap32(spcl.c_flags)
700			    | DR_NEWHEADER);
701		writeheader((ino_t)slp->inode);
702		if (!is_ufs2)
703			spcl.c_flags  = iswap32(iswap32(spcl.c_flags) &
704			    ~ DR_NEWHEADER);
705		msg("Volume %d started at: %s", tapeno, ctime(&tstart_volume));
706		if (tapeno > 1)
707			msg("Volume %d begins with blocks from inode %d\n",
708				tapeno, slp->inode);
709	}
710}
711
712void
713dumpabort(int signo __unused)
714{
715
716	if (master != 0 && master != getpid())
717		/* Signals master to call dumpabort */
718		(void) kill(master, SIGTERM);
719	else {
720#ifdef DUMP_LFS
721		lfs_wrap_go();
722#endif
723		killall();
724		msg("The ENTIRE dump is aborted.\n");
725	}
726#ifdef RDUMP
727	rmtclose();
728#endif
729	Exit(X_ABORT);
730}
731
732void
733Exit(int status)
734{
735
736#ifdef TDEBUG
737	msg("pid = %d exits with status %d\n", getpid(), status);
738#endif /* TDEBUG */
739	exit(status);
740}
741
742/*
743 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
744 */
745static void
746proceed(int signo __unused)
747{
748	caught++;
749}
750
751void
752enslave(void)
753{
754	int cmd[2];
755	int i, j;
756
757	master = getpid();
758
759	signal(SIGTERM, dumpabort);  /* Slave sends SIGTERM on dumpabort() */
760	signal(SIGPIPE, sigpipe);
761	signal(SIGUSR1, tperror);    /* Slave sends SIGUSR1 on tape errors */
762	signal(SIGUSR2, proceed);    /* Slave sends SIGUSR2 to next slave */
763
764	for (i = 0; i < SLAVES; i++) {
765		if (i == slp - &slaves[0]) {
766			caught = 1;
767		} else {
768			caught = 0;
769		}
770
771		if (socketpair(AF_LOCAL, SOCK_STREAM, 0, cmd) < 0 ||
772		    (slaves[i].pid = fork()) < 0)
773			quit("too many slaves, %d (recompile smaller): %s\n",
774			    i, strerror(errno));
775
776		slaves[i].fd = cmd[1];
777		slaves[i].sent = 0;
778		if (slaves[i].pid == 0) { 	    /* Slave starts up here */
779			for (j = 0; j <= i; j++)
780				(void) close(slaves[j].fd);
781			signal(SIGINT, SIG_IGN);    /* Master handles this */
782			signal(SIGINFO, SIG_IGN);
783			doslave(cmd[0], i);
784			Exit(X_FINOK);
785		}
786	}
787
788	for (i = 0; i < SLAVES; i++)
789		(void) atomic_write(slaves[i].fd,
790				(char *) &slaves[(i + 1) % SLAVES].pid,
791				sizeof slaves[0].pid);
792
793	master = 0;
794}
795
796void
797killall(void)
798{
799	int i;
800
801	for (i = 0; i < SLAVES; i++)
802		if (slaves[i].pid > 0) {
803			(void) kill(slaves[i].pid, SIGKILL);
804			slaves[i].sent = 0;
805		}
806}
807
808/*
809 * Synchronization - each process has a lockfile, and shares file
810 * descriptors to the following process's lockfile.  When our write
811 * completes, we release our lock on the following process's lock-
812 * file, allowing the following process to lock it and proceed. We
813 * get the lock back for the next cycle by swapping descriptors.
814 */
815static void
816doslave(int cmd, int slave_number __unused)
817{
818	int nread, nextslave, size, wrote, eot_count, werror;
819	sigset_t nsigset, osigset;
820
821	wrote = 0;
822	/*
823	 * Need our own seek pointer.
824	 */
825	(void) close(diskfd);
826	if ((diskfd = open(disk_dev, O_RDONLY)) < 0)
827		quit("slave couldn't reopen disk: %s\n", strerror(errno));
828
829	/*
830	 * Need the pid of the next slave in the loop...
831	 */
832	if ((nread = atomic_read(cmd, (char *)&nextslave, sizeof nextslave))
833	    != sizeof nextslave) {
834		quit("master/slave protocol botched - didn't get pid of next slave.\n");
835	}
836
837	/*
838	 * Get list of blocks to dump, read the blocks into tape buffer
839	 */
840	while ((nread = atomic_read(cmd, (char *)slp->req, reqsiz)) == reqsiz) {
841		struct req *p = slp->req;
842
843		for (trecno = 0; trecno < ntrec;
844		     trecno += p->count, p += p->count) {
845			if (p->dblk) {
846				bread(p->dblk, slp->tblock[trecno],
847					p->count * TP_BSIZE);
848			} else {
849				if (p->count != 1 || atomic_read(cmd,
850				    (char *)slp->tblock[trecno],
851				    TP_BSIZE) != TP_BSIZE)
852				       quit("master/slave protocol botched.\n");
853			}
854		}
855
856		sigemptyset(&nsigset);
857		sigaddset(&nsigset, SIGUSR2);
858		sigprocmask(SIG_BLOCK, &nsigset, &osigset);
859		while (!caught)
860			sigsuspend(&osigset);
861		caught = 0;
862		sigprocmask(SIG_SETMASK, &osigset, NULL);
863
864		/* Try to write the data... */
865		eot_count = 0;
866		size = 0;
867		werror = 0;
868
869		while (eot_count < 10 && size < writesize) {
870#ifdef RDUMP
871			if (host)
872				wrote = rmtwrite(slp->tblock[0]+size,
873				    writesize-size);
874			else
875#endif
876				wrote = write(tapefd, slp->tblock[0]+size,
877				    writesize-size);
878			werror = errno;
879#ifdef WRITEDEBUG
880			fprintf(stderr, "slave %d wrote %d werror %d\n",
881			    slave_number, wrote, werror);
882#endif
883			if (wrote < 0)
884				break;
885			if (wrote == 0)
886				eot_count++;
887			size += wrote;
888		}
889
890#ifdef WRITEDEBUG
891		if (size != writesize)
892			fprintf(stderr,
893		    "slave %d only wrote %d out of %d bytes and gave up.\n",
894			    slave_number, size, writesize);
895#endif
896
897		/*
898		 * Handle ENOSPC as an EOT condition.
899		 */
900		if (wrote < 0 && werror == ENOSPC) {
901			wrote = 0;
902			eot_count++;
903		}
904
905		if (eot_count > 0)
906			size = 0;
907
908		if (wrote < 0) {
909			(void) kill(master, SIGUSR1);
910			sigemptyset(&nsigset);
911			for (;;)
912				sigsuspend(&nsigset);
913		} else {
914			/*
915			 * pass size of write back to master
916			 * (for EOT handling)
917			 */
918			(void) atomic_write(cmd, (char *)&size, sizeof size);
919		}
920
921		/*
922		 * If partial write, don't want next slave to go.
923		 * Also jolts him awake.
924		 */
925		(void) kill(nextslave, SIGUSR2);
926	}
927	printcachestats();
928	if (nread != 0)
929		quit("error reading command pipe: %s\n", strerror(errno));
930}
931
932/*
933 * Since a read from a pipe may not return all we asked for,
934 * loop until the count is satisfied (or error).
935 */
936static ssize_t
937atomic_read(int fd, char *buf, int count)
938{
939	ssize_t got, need = count;
940
941	while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
942		buf += got;
943	return (got < 0 ? got : count - need);
944}
945
946/*
947 * Since a write may not write all we ask if we get a signal,
948 * loop until the count is satisfied (or error).
949 */
950static ssize_t
951atomic_write(int fd, char *buf, int count)
952{
953	ssize_t got, need = count;
954
955	while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
956		buf += got;
957	return (got < 0 ? got : count - need);
958}
959