Exemple #1
0
/*
 * Acksend implements reliable datagram transmission by using sequence
 * numbers and retransmission when necessary.
 * If `name' is ANYADDR, this routine implements reliable broadcast.
 *
 * Because this function calls readmsg(), none of its args may be in
 *	a message provided by readmsg().
 */
struct tsp *
acksend(struct tsp *message,		/* this message */
	struct sockaddr_in *addr,	/* to here */
	char *name,
	int ack,			/* look for this ack */
	struct netinfo *net,		/* receive from this network */
	int bad)			/* 1=losing patience */
{
	struct timeval twait;
	int count;
	long msec;

	message->tsp_vers = TSPVERSION;
	message->tsp_seq = sequence;
	if (trace) {
		fprintf(fd, "acksend: to %s: ",
			(name == ANYADDR ? "broadcast" : name));
		print(message, addr);
	}
	bytenetorder(message);

	msec = 200;
	count = bad ? 1 : 5;	/* 5 packets in 6.4 seconds */
	answer = NULL;
	do {
		if (!answer) {
			/* do not go crazy transmitting just because the
			 * other guy cannot keep our sequence numbers
			 * straight.
			 */
			if (sendto(sock, (char *)message, sizeof(struct tsp),
				   0, (struct sockaddr*)addr,
				   sizeof(struct sockaddr)) < 0) {
				trace_sendto_err(addr->sin_addr);
				break;
			}
		}

		mstotvround(&twait, msec);
		answer  = readmsg(ack, name, &twait, net);
		if (answer != NULL) {
			if (answer->tsp_seq != sequence) {
				if (trace)
					fprintf(fd,"acksend: seq # %u!=%u\n",
						answer->tsp_seq, sequence);
				continue;
			}
			break;
		}

		msec *= 2;
	} while (--count > 0);
	sequence++;

	return(answer);
}
Exemple #2
0
void
xmit(int type, u_int seq, struct sockaddr_in *addr)
{
	static struct tsp msg;

	msg.tsp_type = type;
	msg.tsp_seq = seq;
	msg.tsp_vers = TSPVERSION;
	(void)strcpy(msg.tsp_name, hostname);
	bytenetorder(&msg);
	if (sendto(sock, (char *)&msg, sizeof(struct tsp), 0,
		   (struct sockaddr*)addr, sizeof(struct sockaddr)) < 0) {
		trace_sendto_err(addr->sin_addr);
	}
}
Exemple #3
0
/*
 * Causes the election timer to expire on the selected hosts
 * It sends just one udp message per machine, relying on
 * reliability of communication channel.
 */
void
testing(int argc, char *argv[])
{
	struct servent *srvp;
	struct sockaddr_in sin;
	struct tsp msg;

	if (argc < 2)  {
		printf("usage: timedc election host1 [host2 ...]\n");
		return;
	}

	srvp = getservbyname("timed", "udp");
	if (srvp == NULL) {
		warnx("timed/udp: unknown service");
		return;
	}

	while (argc > 1) {
		argc--; argv++;
		hp = gethostbyname(*argv);
		if (hp == NULL) {
			warnx("%s: %s", *argv, hstrerror(h_errno));
			argc--; argv++;
			continue;
		}
		sin.sin_port = srvp->s_port;
		sin.sin_family = hp->h_addrtype;
		bcopy(hp->h_addr, &sin.sin_addr.s_addr, hp->h_length);

		msg.tsp_type = TSP_TEST;
		msg.tsp_vers = TSPVERSION;
		if (gethostname(myname, sizeof(myname) - 1) < 0)
			err(1, "gethostname");
		(void)strlcpy(msg.tsp_name, myname, sizeof(msg.tsp_name));
		bytenetorder(&msg);
		if (sendto(sock, &msg, sizeof(struct tsp), 0,
			   (struct sockaddr*)&sin,
			   sizeof(struct sockaddr)) < 0) {
			warn("sendto");
		}
	}
}
Exemple #4
0
/*
 * Enables or disables tracing on local timedaemon
 */
void
tracing(int argc, char *argv[])
{
	int onflag;
	int length;
	ssize_t cc;
	fd_set ready;
	struct sockaddr_in dest;
	struct sockaddr_in from;
	struct timeval tout;
	struct tsp msg;
	struct servent *srvp;

	if (argc != 2) {
		printf("usage: timedc trace { on | off }\n");
		return;
	}

	srvp = getservbyname("timed", "udp");
	if (srvp == NULL) {
		warnx("timed/udp: unknown service");
		return;
	}
	dest.sin_port = srvp->s_port;
	dest.sin_family = AF_INET;

	if (gethostname(myname, sizeof(myname) - 1) < 0)
		err(1, "gethostname");
	hp = gethostbyname(myname);
	bcopy(hp->h_addr, &dest.sin_addr.s_addr, hp->h_length);

	if (strcmp(argv[1], "on") == 0) {
		msg.tsp_type = TSP_TRACEON;
		onflag = ON;
	} else {
		msg.tsp_type = TSP_TRACEOFF;
		onflag = OFF;
	}

	(void)strcpy(msg.tsp_name, myname);
	msg.tsp_vers = TSPVERSION;
	bytenetorder(&msg);
	if (sendto(sock, &msg, sizeof(struct tsp), 0,
		   (struct sockaddr*)&dest, sizeof(struct sockaddr)) < 0) {
		warn("sendto");
		return;
	}

	tout.tv_sec = 5;
	tout.tv_usec = 0;
	FD_ZERO(&ready);
	FD_SET(sock, &ready);
	if (select(FD_SETSIZE, &ready, (fd_set *)0, (fd_set *)0, &tout)) {
		length = sizeof(from);
		cc = recvfrom(sock, &msg, sizeof(struct tsp), 0,
			      (struct sockaddr *)&from, &length);
		if (cc < 0) {
			warn("recvfrom");
			return;
		}
		/*
		 * The 4.3BSD protocol spec had a 32-byte tsp_name field, and
		 * this is still OS-dependent.  Demand that the packet is at
		 * least long enough to hold a 4.3BSD packet.
		 */
		if (cc < (sizeof(struct tsp) - MAXHOSTNAMELEN + 32)) {
			fprintf(stderr, "short packet (%zd/%zu bytes) from %s\n",
			    cc, sizeof(struct tsp) - MAXHOSTNAMELEN + 32,
			    inet_ntoa(from.sin_addr));
			return;
		}
		bytehostorder(&msg);
		if (msg.tsp_type == TSP_ACK)
			if (onflag)
				printf("timed tracing enabled\n");
			else
				printf("timed tracing disabled\n");
		else {
			if (msg.tsp_type >= TSPTYPENUMBER)
				printf("unknown ack received: %u\n",
					msg.tsp_type);
			else	
				printf("wrong ack received: %s\n",
						tsptype[msg.tsp_type]);
		}
	} else
		printf("communication error\n");
}
Exemple #5
0
/*
 * finds location of master timedaemon
 */
void
msite(int argc, char *argv[])
{
	ssize_t cc;
	fd_set ready;
	struct sockaddr_in dest;
	int i, length;
	struct sockaddr_in from;
	struct timeval tout;
	struct tsp msg;
	struct servent *srvp;
	char *tgtname;

	if (argc < 1) {
		printf("usage: timedc msite [host ...]\n");
		return;
	}

	srvp = getservbyname("timed", "udp");
	if (srvp == NULL) {
		warnx("timed/udp: unknown service");
		return;
	}
	dest.sin_port = srvp->s_port;
	dest.sin_family = AF_INET;

	if (gethostname(myname, sizeof(myname) - 1) < 0)
		err(1, "gethostname");
	i = 1;
	do {
		tgtname = (i >= argc) ? myname : argv[i];
		hp = gethostbyname(tgtname);
		if (hp == NULL) {
			warnx("%s: %s", tgtname, hstrerror(h_errno));
			continue;
		}
		bcopy(hp->h_addr, &dest.sin_addr.s_addr, hp->h_length);

		(void)strlcpy(msg.tsp_name, myname, sizeof(msg.tsp_name));
		msg.tsp_type = TSP_MSITE;
		msg.tsp_vers = TSPVERSION;
		bytenetorder(&msg);
		if (sendto(sock, &msg, sizeof(struct tsp), 0,
			   (struct sockaddr*)&dest,
			   sizeof(struct sockaddr)) < 0) {
			warn("sendto");
			continue;
		}

		tout.tv_sec = 15;
		tout.tv_usec = 0;
		FD_ZERO(&ready);
		FD_SET(sock, &ready);
		if (select(FD_SETSIZE, &ready, (fd_set *)0, (fd_set *)0,
			   &tout)) {
			length = sizeof(from);
			cc = recvfrom(sock, &msg, sizeof(struct tsp), 0,
				      (struct sockaddr *)&from, &length);
			if (cc < 0) {
				warn("recvfrom");
				continue;
			}
			/*
			 * The 4.3BSD protocol spec had a 32-byte tsp_name field, and
			 * this is still OS-dependent.  Demand that the packet is at
			 * least long enough to hold a 4.3BSD packet.
			 */
			if (cc < (sizeof(struct tsp) - MAXHOSTNAMELEN + 32)) {
				fprintf(stderr, 
				   "short packet (%zd/%zu bytes) from %s\n",
				   cc, sizeof(struct tsp) - MAXHOSTNAMELEN + 32,
				   inet_ntoa(from.sin_addr));
				continue;
			}
			bytehostorder(&msg);
			if (msg.tsp_type == TSP_ACK) {
				printf("master timedaemon at %s is %s\n",
				       tgtname, msg.tsp_name);
			} else {
				if (msg.tsp_type >= TSPTYPENUMBER)
					printf("unknown ack received: %u\n",
						msg.tsp_type);
				else	
					printf("wrong ack received: %s\n",
				       		tsptype[msg.tsp_type]);
			}
		} else {
			printf("communication error with %s\n", tgtname);
		}
	} while (++i < argc);
}
Exemple #6
0
int
slave()
{
	int tries;
	long electiontime, refusetime, looktime, looptime, adjtime;
	u_short seq;
	long fastelection;
#define FASTTOUT 3
	struct in_addr cadr;
	struct timeval otime;
	struct sockaddr_in taddr;
	char tname[MAXHOSTNAMELEN];
	struct tsp *msg, to;
	struct timeval ntime, wait, tmptv;
	time_t tsp_time_sec;
	struct tsp *answer;
	int timeout();
	char olddate[32];
	char newdate[32];
	struct netinfo *ntp;
	struct hosttbl *htp;
	struct utmpx utx;


	old_slavenet = 0;
	seq = 0;
	refusetime = 0;
	adjtime = 0;

	(void)gettimeofday(&ntime, NULL);
	electiontime = ntime.tv_sec + delay2;
	fastelection = ntime.tv_sec + FASTTOUT;
	if (justquit)
		looktime = electiontime;
	else
		looktime = fastelection;
	looptime = fastelection;

	if (slavenet)
		xmit(TSP_SLAVEUP, 0, &slavenet->dest_addr);
	if (status & MASTER) {
		for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
			if (ntp->status == MASTER)
				masterup(ntp);
		}
	}

loop:
	get_goodgroup(0);
	(void)gettimeofday(&ntime, NULL);
	if (ntime.tv_sec > electiontime) {
		if (trace)
			fprintf(fd, "election timer expired\n");
		longjmp(jmpenv, 1);
	}

	if (ntime.tv_sec >= looktime) {
		if (trace)
			fprintf(fd, "Looking for nets to master\n");

		if (Mflag && nignorednets > 0) {
			for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
				if (ntp->status == IGNORE
				    || ntp->status == NOMASTER) {
					lookformaster(ntp);
					if (ntp->status == MASTER) {
						masterup(ntp);
					} else if (ntp->status == MASTER) {
						ntp->status = NOMASTER;
					}
				}
				if (ntp->status == MASTER
				    && --ntp->quit_count < 0)
					ntp->quit_count = 0;
			}
			makeslave(slavenet);	/* prune extras */
			setstatus();
		}
		(void)gettimeofday(&ntime, NULL);
		looktime = ntime.tv_sec + delay2;
	}
	if (ntime.tv_sec >= looptime) {
		if (trace)
			fprintf(fd, "Looking for loops\n");
		for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
		    if (ntp->status == MASTER) {
			to.tsp_type = TSP_LOOP;
			to.tsp_vers = TSPVERSION;
			to.tsp_seq = sequence++;
			to.tsp_hopcnt = MAX_HOPCNT;
			(void)strcpy(to.tsp_name, hostname);
			bytenetorder(&to);
			if (sendto(sock, (char *)&to, sizeof(struct tsp), 0,
				   (struct sockaddr*)&ntp->dest_addr,
				   sizeof(ntp->dest_addr)) < 0) {
				trace_sendto_err(ntp->dest_addr.sin_addr);
			}
		    }
		}
		(void)gettimeofday(&ntime, NULL);
		looptime = ntime.tv_sec + delay2;
	}

	wait.tv_sec = min(electiontime,min(looktime,looptime)) - ntime.tv_sec;
	if (wait.tv_sec < 0)
		wait.tv_sec = 0;
	wait.tv_sec += FASTTOUT;
	wait.tv_usec = 0;
	msg = readmsg(TSP_ANY, ANYADDR, &wait, 0);

	if (msg != NULL) {
		/*
		 * filter stuff not for us
		 */
		switch (msg->tsp_type) {
		case TSP_SETDATE:
		case TSP_TRACEOFF:
		case TSP_TRACEON:
			/*
			 * XXX check to see they are from ourself
			 */
			break;

		case TSP_TEST:
		case TSP_MSITE:
			break;

		case TSP_MASTERUP:
			if (!fromnet) {
				if (trace) {
					fprintf(fd, "slave ignored: ");
					print(msg, &from);
				}
				goto loop;
			}
			break;

		default:
			if (!fromnet
			    || fromnet->status == IGNORE
			    || fromnet->status == NOMASTER) {
				if (trace) {
					fprintf(fd, "slave ignored: ");
					print(msg, &from);
				}
				goto loop;
			}
			break;
		}


		/*
		 * now process the message
		 */
		switch (msg->tsp_type) {

		case TSP_ADJTIME:
			if (fromnet != slavenet)
				break;
			if (!good_host_name(msg->tsp_name)) {
				syslog(LOG_NOTICE,
				   "attempted time adjustment by %s",
				       msg->tsp_name);
				suppress(&from, msg->tsp_name, fromnet);
				break;
			}
			/*
			 * Speed up loop detection in case we have a loop.
			 * Otherwise the clocks can race until the loop
			 * is found.
			 */
			(void)gettimeofday(&otime, NULL);
			if (adjtime < otime.tv_sec)
				looptime -= (looptime-otime.tv_sec)/2 + 1;

			setmaster(msg);
			if (seq != msg->tsp_seq) {
				seq = msg->tsp_seq;
				synch(tvtomsround(msg->tsp_time));
			}
			(void)gettimeofday(&ntime, NULL);
			electiontime = ntime.tv_sec + delay2;
			fastelection = ntime.tv_sec + FASTTOUT;
			adjtime = ntime.tv_sec + SAMPLEINTVL*2;
			break;

		case TSP_SETTIME:
			if (fromnet != slavenet)
				break;
			if (seq == msg->tsp_seq)
				break;
			seq = msg->tsp_seq;

			/* adjust time for residence on the queue */
			(void)gettimeofday(&otime, NULL);
			adj_msg_time(msg,&otime);
			/*
			 * the following line is necessary due to syslog
			 * calling ctime() which clobbers the static buffer
			 */
			(void)strcpy(olddate, date());
			tsp_time_sec = msg->tsp_time.tv_sec;
			(void)strcpy(newdate, ctime(&tsp_time_sec));

			if (!good_host_name(msg->tsp_name)) {
				syslog(LOG_NOTICE,
			    "attempted time setting by untrusted %s to %s",
				       msg->tsp_name, newdate);
				suppress(&from, msg->tsp_name, fromnet);
				break;
			}

			setmaster(msg);
 			tmptv.tv_sec = msg->tsp_time.tv_sec;
 			tmptv.tv_usec = msg->tsp_time.tv_usec;
			timevalsub(&ntime, &tmptv, &otime);
			if (ntime.tv_sec < MAXADJ && ntime.tv_sec > -MAXADJ) {
				/*
				 * do not change the clock if we can adjust it
				 */
				synch(tvtomsround(ntime));
			} else {
				utx.ut_type = OLD_TIME;
				gettimeofday(&utx.ut_tv, NULL);
				pututxline(&utx);
				(void)settimeofday(&tmptv, 0);
				utx.ut_type = NEW_TIME;
				gettimeofday(&utx.ut_tv, NULL);
				pututxline(&utx);
				syslog(LOG_NOTICE,
				       "date changed by %s from %s",
					msg->tsp_name, olddate);
				if (status & MASTER)
					spreadtime();
			}
			(void)gettimeofday(&ntime, NULL);
			electiontime = ntime.tv_sec + delay2;
			fastelection = ntime.tv_sec + FASTTOUT;

/* This patches a bad protocol bug.  Imagine a system with several networks,
 * where there are a pair of redundant gateways between a pair of networks,
 * each running timed.  Assume that we start with a third machine mastering
 * one of the networks, and one of the gateways mastering the other.
 * Imagine that the third machine goes away and the non-master gateway
 * decides to replace it.  If things are timed just 'right,' we will have
 * each gateway mastering one network for a little while.  If a SETTIME
 * message gets into the network at that time, perhaps from the newly
 * masterful gateway as it was taking control, the SETTIME will loop
 * forever.  Each time a gateway receives it on its slave side, it will
 * call spreadtime to forward it on its mastered network.  We are now in
 * a permanent loop, since the SETTIME msgs will keep any clock
 * in the network from advancing.  Normally, the 'LOOP' stuff will detect
 * and correct the situation.  However, with the clocks stopped, the
 * 'looptime' timer cannot expire.  While they are in this state, the
 * masters will try to saturate the network with SETTIME packets.
 */
			looptime = ntime.tv_sec + (looptime-otime.tv_sec)/2-1;
			break;

		case TSP_MASTERUP:
			if (slavenet && fromnet != slavenet)
				break;
			if (!good_host_name(msg->tsp_name)) {
				suppress(&from, msg->tsp_name, fromnet);
				if (electiontime > fastelection)
					electiontime = fastelection;
				break;
			}
			makeslave(fromnet);
			setmaster(msg);
			setstatus();
			answerdelay();
			xmit(TSP_SLAVEUP, 0, &from);
			(void)gettimeofday(&ntime, NULL);
			electiontime = ntime.tv_sec + delay2;
			fastelection = ntime.tv_sec + FASTTOUT;
			refusetime = 0;
			break;

		case TSP_MASTERREQ:
			if (fromnet->status != SLAVE)
				break;
			(void)gettimeofday(&ntime, NULL);
			electiontime = ntime.tv_sec + delay2;
			break;

		case TSP_SETDATE:
			tsp_time_sec = msg->tsp_time.tv_sec;
			(void)strcpy(newdate, ctime(&tsp_time_sec));
			schgdate(msg, newdate);
			break;

		case TSP_SETDATEREQ:
			if (fromnet->status != MASTER)
				break;
			tsp_time_sec = msg->tsp_time.tv_sec;
			(void)strcpy(newdate, ctime(&tsp_time_sec));
			htp = findhost(msg->tsp_name);
			if (0 == htp) {
				syslog(LOG_WARNING,
				       "DATEREQ from uncontrolled machine");
				break;
			}
			if (!htp->good) {
				syslog(LOG_WARNING,
				"attempted date change by untrusted %s to %s",
				       htp->name, newdate);
				spreadtime();
				break;
			}
			schgdate(msg, newdate);
			break;

		case TSP_TRACEON:
			traceon();
			break;

		case TSP_TRACEOFF:
			traceoff("Tracing ended at %s\n");
			break;

		case TSP_SLAVEUP:
			newslave(msg);
			break;

		case TSP_ELECTION:
			if (fromnet->status == SLAVE) {
				(void)gettimeofday(&ntime, NULL);
				electiontime = ntime.tv_sec + delay2;
				fastelection = ntime.tv_sec + FASTTOUT;
				seq = 0;
				if (!good_host_name(msg->tsp_name)) {
					syslog(LOG_NOTICE,
					       "suppress election of %s",
					       msg->tsp_name);
					to.tsp_type = TSP_QUIT;
					electiontime = fastelection;
				} else if (cadr.s_addr != from.sin_addr.s_addr
					   && ntime.tv_sec < refusetime) {
/* if the candidate has to repeat itself, the old code would refuse it
 * the second time.  That would prevent elections.
 */
					to.tsp_type = TSP_REFUSE;
				} else {
					cadr.s_addr = from.sin_addr.s_addr;
					to.tsp_type = TSP_ACCEPT;
					refusetime = ntime.tv_sec + 30;
				}
				taddr = from;
				(void)strcpy(tname, msg->tsp_name);
				(void)strcpy(to.tsp_name, hostname);
				answerdelay();
				if (!acksend(&to, &taddr, tname,
					     TSP_ACK, 0, 0))
					syslog(LOG_WARNING,
					     "no answer from candidate %s\n",
					       tname);

			} else {	/* fromnet->status == MASTER */
				htp = addmach(msg->tsp_name, &from,fromnet);
				to.tsp_type = TSP_QUIT;
				(void)strcpy(to.tsp_name, hostname);
				if (!acksend(&to, &htp->addr, htp->name,
					     TSP_ACK, 0, htp->noanswer)) {
					syslog(LOG_ERR,
					  "no reply from %s to ELECTION-QUIT",
					       htp->name);
					(void)remmach(htp);
				}
			}
			break;

		case TSP_CONFLICT:
			if (fromnet->status != MASTER)
				break;
			/*
			 * After a network partition, there can be
			 * more than one master: the first slave to
			 * come up will notify here the situation.
			 */
			(void)strcpy(to.tsp_name, hostname);

			/* The other master often gets into the same state,
			 * with boring results.
			 */
			ntp = fromnet;	/* (acksend() can leave fromnet=0 */
			for (tries = 0; tries < 3; tries++) {
				to.tsp_type = TSP_RESOLVE;
				answer = acksend(&to, &ntp->dest_addr,
						 ANYADDR, TSP_MASTERACK,
						 ntp, 0);
				if (answer == NULL)
					break;
				htp = addmach(answer->tsp_name,&from,ntp);
				to.tsp_type = TSP_QUIT;
				answer = acksend(&to, &htp->addr, htp->name,
						 TSP_ACK, 0, htp->noanswer);
				if (!answer) {
					syslog(LOG_WARNING,
				  "conflict error: no reply from %s to QUIT",
						htp->name);
					(void)remmach(htp);
				}
			}
			masterup(ntp);
			break;

		case TSP_MSITE:
			if (!slavenet)
				break;
			taddr = from;
			to.tsp_type = TSP_MSITEREQ;
			to.tsp_vers = TSPVERSION;
			to.tsp_seq = 0;
			(void)strcpy(to.tsp_name, hostname);
			answer = acksend(&to, &slavenet->dest_addr,
					 ANYADDR, TSP_ACK,
					 slavenet, 0);
			if (answer != NULL
			    && good_host_name(answer->tsp_name)) {
				setmaster(answer);
				to.tsp_type = TSP_ACK;
				(void)strcpy(to.tsp_name, answer->tsp_name);
				bytenetorder(&to);
				if (sendto(sock, (char *)&to,
					   sizeof(struct tsp), 0,
					   (struct sockaddr*)&taddr,
					   sizeof(taddr)) < 0) {
					trace_sendto_err(taddr.sin_addr);
				}
			}
			break;

		case TSP_MSITEREQ:
			break;

		case TSP_ACCEPT:
		case TSP_REFUSE:
		case TSP_RESOLVE:
			break;

		case TSP_QUIT:
			doquit(msg);		/* become a slave */
			break;

		case TSP_TEST:
			electiontime = 0;
			break;

		case TSP_LOOP:
			/* looking for loops of masters */
			if (!(status & MASTER))
				break;
			if (fromnet->status == SLAVE) {
			    if (!strcmp(msg->tsp_name, hostname)) {
				/*
				 * Someone forwarded our message back to
				 * us.  There must be a loop.  Tell the
				 * master of this network to quit.
				 *
				 * The other master often gets into
				 * the same state, with boring results.
				 */
				ntp = fromnet;
				for (tries = 0; tries < 3; tries++) {
				    to.tsp_type = TSP_RESOLVE;
				    answer = acksend(&to, &ntp->dest_addr,
						     ANYADDR, TSP_MASTERACK,
						     ntp,0);
				    if (answer == NULL)
					break;
				    taddr = from;
				    (void)strcpy(tname, answer->tsp_name);
				    to.tsp_type = TSP_QUIT;
				    (void)strcpy(to.tsp_name, hostname);
				    if (!acksend(&to, &taddr, tname,
						 TSP_ACK, 0, 1)) {
					syslog(LOG_ERR,
					"no reply from %s to slave LOOP-QUIT",
						 tname);
				    } else {
					electiontime = 0;
				    }
				}
				(void)gettimeofday(&ntime, NULL);
				looptime = ntime.tv_sec + FASTTOUT;
			    } else {
				if (msg->tsp_hopcnt-- < 1)
				    break;
				bytenetorder(msg);
				for (ntp = nettab; ntp != 0; ntp = ntp->next) {
				    if (ntp->status == MASTER
					&& 0 > sendto(sock, (char *)msg,
						      sizeof(struct tsp), 0,
					      (struct sockaddr*)&ntp->dest_addr,
						      sizeof(ntp->dest_addr)))
				    trace_sendto_err(ntp->dest_addr.sin_addr);
				}
			    }
			} else {	/* fromnet->status == MASTER */
			    /*
			     * We should not have received this from a net
			     * we are master on.  There must be two masters,
			     * unless the packet was really from us.
			     */
			    if (from.sin_addr.s_addr
				== fromnet->my_addr.s_addr) {
				if (trace)
				    fprintf(fd,"discarding forwarded LOOP\n");
				break;
			    }

			    /*
			     * The other master often gets into the same
			     * state, with boring results.
			     */
			    ntp = fromnet;
			    for (tries = 0; tries < 3; tries++) {
				to.tsp_type = TSP_RESOLVE;
				answer = acksend(&to, &ntp->dest_addr,
						 ANYADDR, TSP_MASTERACK,
						ntp,0);
				if (!answer)
					break;
				htp = addmach(answer->tsp_name,
					      &from,ntp);
				to.tsp_type = TSP_QUIT;
				(void)strcpy(to.tsp_name, hostname);
				if (!acksend(&to,&htp->addr,htp->name,
					     TSP_ACK, 0, htp->noanswer)) {
					syslog(LOG_ERR,
				    "no reply from %s to master LOOP-QUIT",
					       htp->name);
					(void)remmach(htp);
				}
			    }
			    (void)gettimeofday(&ntime, NULL);
			    looptime = ntime.tv_sec + FASTTOUT;
			}
			break;
		default:
			if (trace) {
				fprintf(fd, "garbage message: ");
				print(msg, &from);
			}
			break;
		}
	}
	goto loop;
}
Exemple #7
0
/*
 * The main function of `master' is to periodically compute the differences
 * (deltas) between its clock and the clocks of the slaves, to compute the
 * network average delta, and to send to the slaves the differences between
 * their individual deltas and the network delta.
 * While waiting, it receives messages from the slaves (i.e. requests for
 * master's name, remote requests to set the network time, ...), and
 * takes the appropriate action.
 */
int
master()
{
	struct hosttbl *htp;
	long pollingtime;
#define POLLRATE 4
	int polls;
	struct timeval wait, ntime;
	time_t tsp_time_sec;
	struct tsp *msg, *answer, to;
	char newdate[32];
	struct sockaddr_in taddr;
	char tname[MAXHOSTNAMELEN];
	struct netinfo *ntp;
	int i;

	syslog(LOG_NOTICE, "This machine is master");
	if (trace)
		fprintf(fd, "This machine is master\n");
	for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
		if (ntp->status == MASTER)
			masterup(ntp);
	}
	(void)gettimeofday(&ntime, 0);
	pollingtime = ntime.tv_sec+3;
	if (justquit)
		polls = 0;
	else
		polls = POLLRATE-1;

/* Process all outstanding messages before spending the long time necessary
 *	to update all timers.
 */
loop:
	(void)gettimeofday(&ntime, 0);
	wait.tv_sec = pollingtime - ntime.tv_sec;
	if (wait.tv_sec < 0)
		wait.tv_sec = 0;
	wait.tv_usec = 0;
	msg = readmsg(TSP_ANY, ANYADDR, &wait, 0);
	if (!msg) {
		(void)gettimeofday(&ntime, 0);
		if (ntime.tv_sec >= pollingtime) {
			pollingtime = ntime.tv_sec + SAMPLEINTVL;
			get_goodgroup(0);

/* If a bogus master told us to quit, we can have decided to ignore a
 * network.  Therefore, periodically try to take over everything.
 */
			polls = (polls + 1) % POLLRATE;
			if (0 == polls && nignorednets > 0) {
				trace_msg("Looking for nets to re-master\n");
				for (ntp = nettab; ntp; ntp = ntp->next) {
					if (ntp->status == IGNORE
					    || ntp->status == NOMASTER) {
						lookformaster(ntp);
						if (ntp->status == MASTER) {
							masterup(ntp);
							polls = POLLRATE-1;
						}
					}
					if (ntp->status == MASTER
					    && --ntp->quit_count < 0)
						ntp->quit_count = 0;
				}
				if (polls != 0)
					setstatus();
			}

			synch(0L);

			for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
				to.tsp_type = TSP_LOOP;
				to.tsp_vers = TSPVERSION;
				to.tsp_seq = sequence++;
				to.tsp_hopcnt = MAX_HOPCNT;
				(void)strcpy(to.tsp_name, hostname);
				bytenetorder(&to);
				if (sendto(sock, (char *)&to,
					   sizeof(struct tsp), 0,
					   (struct sockaddr*)&ntp->dest_addr,
					   sizeof(ntp->dest_addr)) < 0) {
				   trace_sendto_err(ntp->dest_addr.sin_addr);
				}
			}
		}


	} else {
		switch (msg->tsp_type) {

		case TSP_MASTERREQ:
			break;

		case TSP_SLAVEUP:
			newslave(msg);
			break;

		case TSP_SETDATE:
			/*
			 * XXX check to see it is from ourself
			 */
			tsp_time_sec = msg->tsp_time.tv_sec;
			(void)strcpy(newdate, ctime(&tsp_time_sec));
			if (!good_host_name(msg->tsp_name)) {
				syslog(LOG_NOTICE,
				       "attempted date change by %s to %s",
				       msg->tsp_name, newdate);
				spreadtime();
				break;
			}

			mchgdate(msg);
			(void)gettimeofday(&ntime, 0);
			pollingtime = ntime.tv_sec + SAMPLEINTVL;
			break;

		case TSP_SETDATEREQ:
			if (!fromnet || fromnet->status != MASTER)
				break;
			tsp_time_sec = msg->tsp_time.tv_sec;
			(void)strcpy(newdate, ctime(&tsp_time_sec));
			htp = findhost(msg->tsp_name);
			if (htp == 0) {
				syslog(LOG_ERR,
				       "attempted SET DATEREQ by uncontrolled %s to %s",
				       msg->tsp_name, newdate);
				break;
			}
			if (htp->seq == msg->tsp_seq)
				break;
			htp->seq = msg->tsp_seq;
			if (!htp->good) {
				syslog(LOG_NOTICE,
				"attempted SET DATEREQ by untrusted %s to %s",
				       msg->tsp_name, newdate);
				spreadtime();
				break;
			}

			mchgdate(msg);
			(void)gettimeofday(&ntime, 0);
			pollingtime = ntime.tv_sec + SAMPLEINTVL;
			break;

		case TSP_MSITE:
			xmit(TSP_ACK, msg->tsp_seq, &from);
			break;

		case TSP_MSITEREQ:
			break;

		case TSP_TRACEON:
			traceon();
			break;

		case TSP_TRACEOFF:
			traceoff("Tracing ended at %s\n");
			break;

		case TSP_ELECTION:
			if (!fromnet)
				break;
			if (fromnet->status == MASTER) {
				pollingtime = 0;
				(void)addmach(msg->tsp_name, &from,fromnet);
			}
			taddr = from;
			(void)strcpy(tname, msg->tsp_name);
			to.tsp_type = TSP_QUIT;
			(void)strcpy(to.tsp_name, hostname);
			answer = acksend(&to, &taddr, tname,
					 TSP_ACK, 0, 1);
			if (answer == NULL) {
				syslog(LOG_ERR, "election error by %s",
				       tname);
			}
			break;

		case TSP_CONFLICT:
			/*
			 * After a network partition, there can be
			 * more than one master: the first slave to
			 * come up will notify here the situation.
			 */
			if (!fromnet || fromnet->status != MASTER)
				break;
			(void)strcpy(to.tsp_name, hostname);

			/* The other master often gets into the same state,
			 * with boring results if we stay at it forever.
			 */
			ntp = fromnet;	/* (acksend() can leave fromnet=0 */
			for (i = 0; i < 3; i++) {
				to.tsp_type = TSP_RESOLVE;
				(void)strcpy(to.tsp_name, hostname);
				answer = acksend(&to, &ntp->dest_addr,
						 ANYADDR, TSP_MASTERACK,
						 ntp, 0);
				if (!answer)
					break;
				htp = addmach(answer->tsp_name,&from,ntp);
				to.tsp_type = TSP_QUIT;
				msg = acksend(&to, &htp->addr, htp->name,
					      TSP_ACK, 0, htp->noanswer);
				if (msg == NULL) {
					syslog(LOG_ERR,
				    "no response from %s to CONFLICT-QUIT",
					       htp->name);
				}
			}
			masterup(ntp);
			pollingtime = 0;
			break;

		case TSP_RESOLVE:
			if (!fromnet || fromnet->status != MASTER)
				break;
			/*
			 * do not want to call synch() while waiting
			 * to be killed!
			 */
			(void)gettimeofday(&ntime, (struct timezone *)0);
			pollingtime = ntime.tv_sec + SAMPLEINTVL;
			break;

		case TSP_QUIT:
			doquit(msg);		/* become a slave */
			break;

		case TSP_LOOP:
			if (!fromnet || fromnet->status != MASTER
			    || !strcmp(msg->tsp_name, hostname))
				break;
			/*
			 * We should not have received this from a net
			 * we are master on.  There must be two masters.
			 */
			htp = addmach(msg->tsp_name, &from,fromnet);
			to.tsp_type = TSP_QUIT;
			(void)strcpy(to.tsp_name, hostname);
			answer = acksend(&to, &htp->addr, htp->name,
					 TSP_ACK, 0, 1);
			if (!answer) {
				syslog(LOG_WARNING,
				"loop breakage: no reply from %s=%s to QUIT",
				    htp->name, inet_ntoa(htp->addr.sin_addr));
				(void)remmach(htp);
			}

		case TSP_TEST:
			if (trace) {
				fprintf(fd,
		"\tnets = %d, masters = %d, slaves = %d, ignored = %d\n",
		nnets, nmasternets, nslavenets, nignorednets);
				setstatus();
			}
			pollingtime = 0;
			polls = POLLRATE-1;
			break;

		default:
			if (trace) {
				fprintf(fd, "garbage message: ");
				print(msg, &from);
			}
			break;
		}
	}
	goto loop;
}
Exemple #8
0
slave()
{
	int length;
	int senddateack;
	long electiontime, refusetime, looktime;
	u_short seq;
	char candidate[MAXHOSTNAMELEN];
	struct tsp *msg, to, *readmsg();
	struct sockaddr_in saveaddr, msaveaddr;
	struct timeval wait;
	struct timeval time, otime;
	struct tsp *answer, *acksend();
	int timeout();
	char *date();
	long casual();
	int bytenetorder();
	char olddate[32];
	struct sockaddr_in server;
	register struct netinfo *ntp;
	int ind;
	struct tsp resp;
	extern int Mflag;
	extern int justquit;
#ifdef MEASURE
	extern FILE *fp;
#endif
	if (slavenet) {
		resp.tsp_type = TSP_SLAVEUP;
		resp.tsp_vers = TSPVERSION;
		(void)strcpy(resp.tsp_name, hostname);
		bytenetorder(&resp);
		if (sendto(sock, (char *)&resp, sizeof(struct tsp), 0,
		    &slavenet->dest_addr, sizeof(struct sockaddr_in)) < 0) {
			syslog(LOG_ERR, "sendto: %m");
			exit(1);
		}
	}

	if (status & MASTER) {
#ifdef MEASURE
		if (fp == NULL) {
			fp = fopen("/usr/adm/timed.masterlog", "w");
			setlinebuf(fp);
		}
#endif
		syslog(LOG_INFO, "THIS MACHINE IS A SUBMASTER");
		if (trace) {
			fprintf(fd, "THIS MACHINE IS A SUBMASTER\n");
		}
		for (ntp = nettab; ntp != NULL; ntp = ntp->next)
			if (ntp->status == MASTER)
				masterup(ntp);

	} else {
		syslog(LOG_INFO, "THIS MACHINE IS A SLAVE");
		if (trace) {
			fprintf(fd, "THIS MACHINE IS A SLAVE\n");
		}
	}

	seq = 0;
	senddateack = OFF;
	refusetime = 0;

	(void)gettimeofday(&time, (struct timezone *)0);
	electiontime = time.tv_sec + delay2;
	if (Mflag)
		if (justquit)
			looktime = time.tv_sec + delay2;
		else 
			looktime = 1;
	else
		looktime = 0;

loop:
	length = sizeof(struct sockaddr_in);
	(void)gettimeofday(&time, (struct timezone *)0);
	if (time.tv_sec > electiontime) {
		if (trace) 
			fprintf(fd, "election timer expired\n");
		longjmp(jmpenv, 1);
	}
	if (looktime && time.tv_sec > looktime) {
		if (trace) 
			fprintf(fd, "Looking for nets to master and loops\n");
		
		if (nignorednets > 0) {
			for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
				if (ntp->status == IGNORE) {
					lookformaster(ntp);
					if (ntp->status == MASTER)
						masterup(ntp);
					else
						ntp->status = IGNORE;
				}
			}
			setstatus();
#ifdef MEASURE
			/*
			 * Check to see if we just became master
			 * (file not open)
			 */
			if (fp == NULL) {
				fp = fopen("/usr/adm/timed.masterlog", "w");
				setlinebuf(fp);
			}
#endif
		}

		for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
		    if (ntp->status == MASTER) {
			to.tsp_type = TSP_LOOP;
			to.tsp_vers = TSPVERSION;
			to.tsp_seq = sequence++;
			to.tsp_hopcnt = 10;
			(void)strcpy(to.tsp_name, hostname);
			bytenetorder(&to);
			if (sendto(sock, (char *)&to, sizeof(struct tsp), 0,
			    &ntp->dest_addr, sizeof(struct sockaddr_in)) < 0) {
				syslog(LOG_ERR, "sendto: %m");
				exit(1);
			}
		    }
		}
		(void)gettimeofday(&time, (struct timezone *)0);
		looktime = time.tv_sec + delay2;
	}
	wait.tv_sec = electiontime - time.tv_sec + 10;
	wait.tv_usec = 0;
	msg = readmsg(TSP_ANY, (char *)ANYADDR, &wait, (struct netinfo *)NULL);
	if (msg != NULL) {
		switch (msg->tsp_type) {
		case TSP_SETDATE:
#ifdef TESTING
		case TSP_TEST:
#endif
		case TSP_MSITE:
		case TSP_TRACEOFF:
		case TSP_TRACEON:
			break;
		case TSP_MASTERUP:
			if (fromnet == NULL) {
				if (trace) {
					fprintf(fd, "slave ignored: ");
					print(msg, &from);
				}
				goto loop;
			}
			break;
		default:
			if (fromnet == NULL || fromnet->status == IGNORE) {
				if (trace) {
					fprintf(fd, "slave ignored: ");
					print(msg, &from);
				}
				goto loop;
			}
			break;
		}

		switch (msg->tsp_type) {

		case TSP_ADJTIME:
			if (fromnet->status != SLAVE)
				break;
			(void)gettimeofday(&time, (struct timezone *)0);
			electiontime = time.tv_sec + delay2;
			if (seq != msg->tsp_seq) {
				seq = msg->tsp_seq;
				if ((status & SUBMASTER) == SUBMASTER) {
					synch((msg->tsp_time.tv_sec * 1000) + 
					    (msg->tsp_time.tv_usec / 1000));
				} else {
					adjclock(&(msg->tsp_time));
				}
			}
			break;
		case TSP_SETTIME:
			if (fromnet->status != SLAVE)
				break;
			if (seq == msg->tsp_seq)
				break;

			seq = msg->tsp_seq;

			(void)strcpy(olddate, date());
			(void)gettimeofday(&otime, (struct timezone *)0);
			(void)settimeofday(&msg->tsp_time,
				(struct timezone *)0);
			syslog(LOG_NOTICE, "date changed by %s from: %s",
				msg->tsp_name, olddate);
			logwtmp(otime, msg->tsp_time);
			if ((status & SUBMASTER) == SUBMASTER)
				spreadtime();
			(void)gettimeofday(&time, (struct timezone *)0);
			electiontime = time.tv_sec + delay2;

			if (senddateack == ON) {
				senddateack = OFF;
				msg->tsp_type = TSP_DATEACK;
				(void)strcpy(msg->tsp_name, hostname);
				bytenetorder(msg);
				length = sizeof(struct sockaddr_in);
				if (sendto(sock, (char *)msg, 
						sizeof(struct tsp), 0,
						&saveaddr, length) < 0) {
					syslog(LOG_ERR, "sendto: %m");
					exit(1);
				}
			}
			break;
		case TSP_MASTERUP:
			if (slavenet && fromnet != slavenet)
				break;
			makeslave(fromnet);
			setstatus();
			msg->tsp_type = TSP_SLAVEUP;
			msg->tsp_vers = TSPVERSION;
			(void)strcpy(msg->tsp_name, hostname);
			bytenetorder(msg);
			answerdelay();
			length = sizeof(struct sockaddr_in);
			if (sendto(sock, (char *)msg, sizeof(struct tsp), 0, 
						&from, length) < 0) {
				syslog(LOG_ERR, "sendto: %m");
				exit(1);
			}
			backoff = 1;
			delay2 = casual((long)MINTOUT, (long)MAXTOUT);
			(void)gettimeofday(&time, (struct timezone *)0);
			electiontime = time.tv_sec + delay2;
			refusetime = 0;
			break;
		case TSP_MASTERREQ:
			if (fromnet->status != SLAVE)
				break;
			(void)gettimeofday(&time, (struct timezone *)0);
			electiontime = time.tv_sec + delay2;
			break;
		case TSP_SETDATE:
			saveaddr = from;
			msg->tsp_type = TSP_SETDATEREQ;
			msg->tsp_vers = TSPVERSION;
			(void)strcpy(msg->tsp_name, hostname);
			for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
				if (ntp->status == SLAVE)
					break;
			}
			if (ntp == NULL)
				break;
			answer = acksend(msg, &ntp->dest_addr, (char *)ANYADDR,
			    TSP_DATEACK, ntp);
			if (answer != NULL) {
				msg->tsp_type = TSP_ACK;
				bytenetorder(msg);
				length = sizeof(struct sockaddr_in);
				if (sendto(sock, (char *)msg,
				    sizeof(struct tsp), 0, &saveaddr,
				    length) < 0) {
					syslog(LOG_ERR, "sendto: %m");
					exit(1);
				}
				senddateack = ON;
			}
			break;
		case TSP_SETDATEREQ:
			saveaddr = from;
			if (status != SUBMASTER || fromnet->status != MASTER)
				break;
			for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
				if (ntp->status == SLAVE)
					break;
			}
			ind = findhost(msg->tsp_name);
			if (ind < 0) {
			    syslog(LOG_WARNING,
				"DATEREQ from uncontrolled machine");
			    break;
			}
			syslog(LOG_DEBUG,
			    "forwarding date change request for %s",
			    msg->tsp_name);
			(void)strcpy(msg->tsp_name, hostname);
			answer = acksend(msg, &ntp->dest_addr, (char *)ANYADDR,
			    TSP_DATEACK, ntp);
			if (answer != NULL) {
				msg->tsp_type = TSP_DATEACK;
				bytenetorder(msg);
				length = sizeof(struct sockaddr_in);
				if (sendto(sock, (char *)msg,
				    sizeof(struct tsp), 0, &saveaddr,
				    length) < 0) {
					syslog(LOG_ERR, "sendto: %m");
					exit(1);
				}
			}
			break;
		case TSP_TRACEON:
			if (!(trace)) {
				fd = fopen(tracefile, "w");
				setlinebuf(fd);
				fprintf(fd, "Tracing started on: %s\n\n", 
								date());
			}
			trace = ON;
			break;
		case TSP_TRACEOFF:
			if (trace) {
				fprintf(fd, "Tracing ended on: %s\n", date());
				(void)fclose(fd);
			}
#ifdef GPROF
			moncontrol(0);
			_mcleanup();
			moncontrol(1);
#endif
			trace = OFF;
			break;
		case TSP_SLAVEUP:
			if ((status & MASTER) && fromnet->status == MASTER) {
				ind = addmach(msg->tsp_name, &from);
				newslave(ind, msg->tsp_seq);
			}
			break;
		case TSP_ELECTION:
			if (fromnet->status == SLAVE) {
				(void)gettimeofday(&time, (struct timezone *)0);
				electiontime = time.tv_sec + delay2;
				seq = 0;            /* reset sequence number */
				if (time.tv_sec < refusetime)
					msg->tsp_type = TSP_REFUSE;
				else {
					msg->tsp_type = TSP_ACCEPT;
					refusetime = time.tv_sec + 30;
				}
				(void)strcpy(candidate, msg->tsp_name);
				(void)strcpy(msg->tsp_name, hostname);
				answerdelay();
				server = from;
				answer = acksend(msg, &server, candidate, TSP_ACK,
				    (struct netinfo *)NULL);
				if (answer == NULL)
					syslog(LOG_WARNING,
					   "no answer from master candidate\n");
			} else {	/* fromnet->status == MASTER */
				to.tsp_type = TSP_QUIT;
				(void)strcpy(to.tsp_name, hostname);
				server = from;
				answer = acksend(&to, &server, msg->tsp_name,
				    TSP_ACK, (struct netinfo *)NULL);
				if (answer == NULL) {
					syslog(LOG_WARNING,
					    "election error: no reply to QUIT");
				} else {
					(void) addmach(msg->tsp_name, &from);
				}
			}
			break;
                case TSP_CONFLICT:
			if (fromnet->status != MASTER)
				break;
                        /*
                         * After a network partition, there can be
                         * more than one master: the first slave to
                         * come up will notify here the situation.
                         */
                        (void)strcpy(to.tsp_name, hostname);

                        if (fromnet == NULL)
                                break;
                        for(;;) {
                                to.tsp_type = TSP_RESOLVE;
                                answer = acksend(&to, &fromnet->dest_addr,
                                    (char *)ANYADDR, TSP_MASTERACK, fromnet);
                                if (answer == NULL)
                                        break;
                                to.tsp_type = TSP_QUIT;
                                server = from;
                                msg = acksend(&to, &server, answer->tsp_name,
                                    TSP_ACK, (struct netinfo *)NULL);
                                if (msg == NULL) {
                                        syslog(LOG_WARNING,
					    "conflict error: no reply to QUIT");
				} else {
                                        (void) addmach(answer->tsp_name, &from);
				}
                        }
                        masterup(fromnet);
                        break;
		case TSP_MSITE:
			if (!slavenet)
				break;
			msaveaddr = from;
			msg->tsp_type = TSP_MSITEREQ;
			msg->tsp_vers = TSPVERSION;
			(void)strcpy(msg->tsp_name, hostname);
			answer = acksend(msg, &slavenet->dest_addr,
					 (char *)ANYADDR, TSP_ACK, slavenet);
			if (answer != NULL) {
				msg->tsp_type = TSP_ACK;
				length = sizeof(struct sockaddr_in);
				bytenetorder(msg);
				if (sendto(sock, (char *)msg, 
						sizeof(struct tsp), 0,
						&msaveaddr, length) < 0) {
					syslog(LOG_ERR, "sendto: %m");
					exit(1);
				}
			}
			break;
		case TSP_ACCEPT:
		case TSP_REFUSE:
			break;
		case TSP_RESOLVE:
			break;
		case TSP_QUIT:
			/* become slave */
#ifdef MEASURE
			if (fp != NULL) {
				(void)fclose(fp);
				fp = NULL;
			}
#endif
			longjmp(jmpenv, 2);
			break;
#ifdef TESTING
		case TSP_TEST:
			electiontime = 0;
			break;
#endif
		case TSP_MSITEREQ:
			if (status & MASTER)
				break;
			if (trace) {
				fprintf(fd, "garbage: ");
				print(msg, &from);
			}
			break;

		case TSP_LOOP:
			/* looking for loops of masters */
			if ( !(status & MASTER))
				break;
			if (fromnet->status == SLAVE) {
			    if ( !strcmp(msg->tsp_name, hostname)) {
				  for(;;) {
				    to.tsp_type = TSP_RESOLVE;
				    answer = acksend(&to, &fromnet->dest_addr,
					(char *)ANYADDR, TSP_MASTERACK,
					fromnet);
				    if (answer == NULL)
					    break;
				    to.tsp_type = TSP_QUIT;
				    (void)strcpy(to.tsp_name, hostname);
				    server = from;
				    answer = acksend(&to, &server,
					answer->tsp_name, TSP_ACK,
					(struct netinfo *)NULL);
				    if (answer == NULL) {
					syslog(LOG_ERR, "loop kill error");
				    } else {
					electiontime = 0;
				    }
				  }
			    } else {
				if (msg->tsp_hopcnt-- <= 0)
				    break;
				bytenetorder(msg);
				ntp = nettab;
				for (; ntp != NULL; ntp = ntp->next)
				    if (ntp->status == MASTER)
					if (sendto(sock, (char *)msg, 
					    sizeof(struct tsp), 0,
					    &ntp->dest_addr, length) < 0) {
						syslog(LOG_ERR, "sendto: %m");
						exit(1);
					}
			    }
			} else {
			    /*
			     * We should not have received this from a net
			     * we are master on.  There must be two masters
			     * in this case.
			     */
			    if (fromnet->my_addr.s_addr == from.sin_addr.s_addr)
				break;
			    for (;;) {
				to.tsp_type = TSP_RESOLVE;
				answer = acksend(&to, &fromnet->dest_addr,
				    (char *)ANYADDR, TSP_MASTERACK,
				    fromnet);
				if (answer == NULL)
					break;
				to.tsp_type = TSP_QUIT;
				(void)strcpy(to.tsp_name, hostname);
				server = from;
				answer = acksend(&to, &server, answer->tsp_name,
				    TSP_ACK, (struct netinfo *)NULL);
				if (answer == NULL) {
					syslog(LOG_ERR, "loop kill error2");
				} else {
					(void)addmach(msg->tsp_name, &from);
				}
			    }
			}
			break;
		default:
			if (trace) {
				fprintf(fd, "garbage: ");
				print(msg, &from);
			}
			break;
		}
	}
	goto loop;
}
Exemple #9
0
master()
{
	int ind;
	long pollingtime;
	struct timeval wait;
	struct timeval time;
	struct timezone tzone;
	struct tsp *msg, to;
	struct sockaddr_in saveaddr;
	int findhost();
	char *date();
	struct tsp *readmsg();
	struct tsp *answer, *acksend();
	char olddate[32];
	struct sockaddr_in server;
	register struct netinfo *ntp;

#ifdef MEASURE
	if (fp == NULL) {
		fp = fopen(_PATH_MASTERLOG, "w");
		setlinebuf(fp);
	}
#endif

	syslog(LOG_INFO, "This machine is master");
	if (trace)
		fprintf(fd, "THIS MACHINE IS MASTER\n");

	for (ntp = nettab; ntp != NULL; ntp = ntp->next)
		if (ntp->status == MASTER)
			masterup(ntp);
	pollingtime = 0;

loop:
	(void)gettimeofday(&time, (struct timezone *)0);
	if (time.tv_sec >= pollingtime) {
		pollingtime = time.tv_sec + SAMPLEINTVL;
		synch(0L);

		for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
			to.tsp_type = TSP_LOOP;
			to.tsp_vers = TSPVERSION;
			to.tsp_seq = sequence++;
			to.tsp_hopcnt = 10;
			(void)strcpy(to.tsp_name, hostname);
			bytenetorder(&to);
			if (sendto(sock, (char *)&to, sizeof(struct tsp), 0,
			    (struct sockaddr *)&ntp->dest_addr,
			    sizeof(struct sockaddr_in)) < 0) {
				syslog(LOG_ERR, "sendto: %m");
				exit(1);
			}
		}
	}

	wait.tv_sec = pollingtime - time.tv_sec;
	wait.tv_usec = 0;
	msg = readmsg(TSP_ANY, (char *)ANYADDR, &wait, (struct netinfo *)NULL);
	if (msg != NULL) {
		switch (msg->tsp_type) {

		case TSP_MASTERREQ:
			break;
		case TSP_SLAVEUP:
			ind = addmach(msg->tsp_name, &from);
			newslave(ind, msg->tsp_seq);
			break;
		case TSP_SETDATE:
			saveaddr = from;
			/*
			 * the following line is necessary due to syslog
			 * calling ctime() which clobbers the static buffer
			 */
			(void)strcpy(olddate, date());
			(void)gettimeofday(&time, &tzone);
			time.tv_sec = msg->tsp_time.tv_sec;
			time.tv_usec = msg->tsp_time.tv_usec;
			logwtmp("|", "date", "");
			(void)settimeofday(&time, &tzone);
			logwtmp("}", "date", "");
			syslog(LOG_NOTICE, "date changed from: %s", olddate);
			msg->tsp_type = TSP_DATEACK;
			msg->tsp_vers = TSPVERSION;
			(void)strcpy(msg->tsp_name, hostname);
			bytenetorder(msg);
			if (sendto(sock, (char *)msg, sizeof(struct tsp), 0,
			    (struct sockaddr *)&saveaddr,
			    sizeof(struct sockaddr_in)) < 0) {
				syslog(LOG_ERR, "sendto: %m");
				exit(1);
			}
			spreadtime();
			pollingtime = 0;
			break;
		case TSP_SETDATEREQ:
			ind = findhost(msg->tsp_name);
			if (ind < 0) { 
			    syslog(LOG_WARNING,
				"DATEREQ from uncontrolled machine");
			    break;
			}
			if (hp[ind].seq !=  msg->tsp_seq) {
				hp[ind].seq = msg->tsp_seq;
				/*
				 * the following line is necessary due to syslog
				 * calling ctime() which clobbers the static buffer
				 */
				(void)strcpy(olddate, date());
				(void)gettimeofday(&time, &tzone);
				time.tv_sec = msg->tsp_time.tv_sec;
				time.tv_usec = msg->tsp_time.tv_usec;
				logwtmp("|", "date", "");
				(void)settimeofday(&time, &tzone);
				logwtmp("{", "date", "");
				syslog(LOG_NOTICE,
				    "date changed by %s from: %s",
				    msg->tsp_name, olddate);
				spreadtime();
				pollingtime = 0;
			}
			break;
		case TSP_MSITE:
		case TSP_MSITEREQ:
			break;
		case TSP_TRACEON:
			if (!(trace)) {
				fd = fopen(tracefile, "w");
				setlinebuf(fd);
				fprintf(fd, "Tracing started on: %s\n\n", 
							date());
			}
			trace = ON;
			break;
		case TSP_TRACEOFF:
			if (trace) {
				fprintf(fd, "Tracing ended on: %s\n", date());
				(void)fclose(fd);
			}
#ifdef GPROF
			moncontrol(0);
			_mcleanup();
			moncontrol(1);
#endif
			trace = OFF;
			break;
		case TSP_ELECTION:
			to.tsp_type = TSP_QUIT;
			(void)strcpy(to.tsp_name, hostname);
			server = from;
			answer = acksend(&to, &server, msg->tsp_name, TSP_ACK,
			    (struct netinfo *)NULL);
			if (answer == NULL) {
				syslog(LOG_ERR, "election error");
			} else {
				(void) addmach(msg->tsp_name, &from);
			}
			pollingtime = 0;
			break;
		case TSP_CONFLICT:
			/*
			 * After a network partition, there can be 
			 * more than one master: the first slave to 
			 * come up will notify here the situation.
			 */

			(void)strcpy(to.tsp_name, hostname);

			if (fromnet == NULL)
				break;
			for(;;) {
				to.tsp_type = TSP_RESOLVE;
				answer = acksend(&to, &fromnet->dest_addr,
				    (char *)ANYADDR, TSP_MASTERACK, fromnet);
				if (answer == NULL)
					break;
				to.tsp_type = TSP_QUIT;
				server = from;
				msg = acksend(&to, &server, answer->tsp_name,
				    TSP_ACK, (struct netinfo *)NULL);
				if (msg == NULL) {
					syslog(LOG_ERR, "error on sending QUIT");
				} else {
					(void) addmach(answer->tsp_name, &from);
				}
			}
			masterup(fromnet);
			pollingtime = 0;
			break;
		case TSP_RESOLVE:
			/*
			 * do not want to call synch() while waiting
			 * to be killed!
			 */
			(void)gettimeofday(&time, (struct timezone *)0);
			pollingtime = time.tv_sec + SAMPLEINTVL;
			break;
		case TSP_QUIT:
			/* become slave */
#ifdef MEASURE
			if (fp != NULL) {
				(void)fclose(fp);
				fp = NULL;
			}
#endif
			longjmp(jmpenv, 2);
			break;
		case TSP_LOOP:
			/*
			 * We should not have received this from a net
			 * we are master on.  There must be two masters
			 * in this case.
			 */
			to.tsp_type = TSP_QUIT;
			(void)strcpy(to.tsp_name, hostname);
			server = from;
			answer = acksend(&to, &server, msg->tsp_name, TSP_ACK,
				(struct netinfo *)NULL);
			if (answer == NULL) {
				syslog(LOG_WARNING,
				    "loop breakage: no reply to QUIT");
			} else {
				(void)addmach(msg->tsp_name, &from);
			}
		default:
			if (trace) {
				fprintf(fd, "garbage: ");
				print(msg, &from);
			}
			break;
		}
	}
	goto loop;
}
Exemple #10
0
/*
 * `election' candidates a host as master: it is called by a slave
 * which runs with the -M option set when its election timeout expires.
 * Note the conservative approach: if a new timed comes up, or another
 * candidate sends an election request, the candidature is withdrawn.
 */
int
election(struct netinfo *net)
{
	struct tsp *resp, msg;
	struct timeval then, wait;
	struct tsp *answer;
	struct hosttbl *htp;
	char loop_lim = 0;

/* This code can get totally confused if it gets slightly behind.  For
 *	example, if readmsg() has some QUIT messages waiting from the last
 *	round, we would send an ELECTION message, get the stale QUIT,
 *	and give up.  This results in network storms when several machines
 *	do it at once.
 */
	wait.tv_sec = 0;
	wait.tv_usec = 0;
	while (0 != readmsg(TSP_REFUSE, ANYADDR, &wait, net)) {
		if (trace)
			fprintf(fd, "election: discarded stale REFUSE\n");
	}
	while (0 != readmsg(TSP_QUIT, ANYADDR, &wait, net)) {
		if (trace)
			fprintf(fd, "election: discarded stale QUIT\n");
	}

again:
	syslog(LOG_INFO, "This machine is a candidate time master");
	if (trace)
		fprintf(fd, "This machine is a candidate time master\n");
	msg.tsp_type = TSP_ELECTION;
	msg.tsp_vers = TSPVERSION;
	(void)strcpy(msg.tsp_name, hostname);
	bytenetorder(&msg);
	if (sendto(sock, (char *)&msg, sizeof(struct tsp), 0,
		   (struct sockaddr*)&net->dest_addr,
		   sizeof(struct sockaddr)) < 0) {
		trace_sendto_err(net->dest_addr.sin_addr);
		return(SLAVE);
	}

	(void)gettimeofday(&then, 0);
	then.tv_sec += 3;
	for (;;) {
		(void)gettimeofday(&wait, 0);
		timevalsub(&wait,&then,&wait);
		resp = readmsg(TSP_ANY, ANYADDR, &wait, net);
		if (!resp)
			return(MASTER);

		switch (resp->tsp_type) {

		case TSP_ACCEPT:
			(void)addmach(resp->tsp_name, &from,fromnet);
			break;

		case TSP_MASTERUP:
		case TSP_MASTERREQ:
			/*
			 * If another timedaemon is coming up at the same
			 * time, give up, and let it be the master.
			 */
			if (++loop_lim < 5
			    && !good_host_name(resp->tsp_name)) {
				(void)addmach(resp->tsp_name, &from,fromnet);
				suppress(&from, resp->tsp_name, net);
				goto again;
			}
			rmnetmachs(net);
			return(SLAVE);

		case TSP_QUIT:
		case TSP_REFUSE:
			/*
			 * Collision: change value of election timer
			 * using exponential backoff.
			 *
			 *  Fooey.
			 * An exponential backoff on a delay starting at
			 * 6 to 15 minutes for a process that takes
			 * milliseconds is silly.  It is particularly
			 * strange that the original code would increase
			 * the backoff without bound.
			 */
			rmnetmachs(net);
			return(SLAVE);

		case TSP_ELECTION:
			/* no master for another round */
			htp = addmach(resp->tsp_name,&from,fromnet);
			msg.tsp_type = TSP_REFUSE;
			(void)strcpy(msg.tsp_name, hostname);
			answer = acksend(&msg, &htp->addr, htp->name,
					 TSP_ACK, 0, htp->noanswer);
			if (!answer) {
				syslog(LOG_ERR, "error in election from %s",
				       htp->name);
			}
			break;

		case TSP_SLAVEUP:
			(void)addmach(resp->tsp_name, &from,fromnet);
			break;

		case TSP_SETDATE:
		case TSP_SETDATEREQ:
			break;

		default:
			if (trace) {
				fprintf(fd, "candidate: ");
				print(resp, &from);
			}
			break;
		}
	}
}