2 * Copyright (c) 1985, 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
32 static char sccsid[] = "@(#)master.c 8.1 (Berkeley) 6/6/93";
34 static const char rcsid[] =
40 #include <sys/types.h>
41 #include <sys/times.h>
44 #include "pathnames.h"
46 extern int measure_delta;
47 extern jmp_buf jmpenv;
52 static int slvcount; /* slaves listening to our clock */
54 static void mchgdate(struct tsp *);
57 * The main function of `master' is to periodically compute the differences
58 * (deltas) between its clock and the clocks of the slaves, to compute the
59 * network average delta, and to send to the slaves the differences between
60 * their individual deltas and the network delta.
61 * While waiting, it receives messages from the slaves (i.e. requests for
62 * master's name, remote requests to set the network time, ...), and
63 * takes the appropriate action.
72 struct timeval wait, ntime;
74 struct tsp *msg, *answer, to;
76 struct sockaddr_in taddr;
77 char tname[MAXHOSTNAMELEN];
81 syslog(LOG_NOTICE, "This machine is master");
83 fprintf(fd, "This machine is master\n");
84 for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
85 if (ntp->status == MASTER)
88 (void)gettimeofday(&ntime, NULL);
89 pollingtime = ntime.tv_sec+3;
95 /* Process all outstanding messages before spending the long time necessary
96 * to update all timers.
99 (void)gettimeofday(&ntime, NULL);
100 wait.tv_sec = pollingtime - ntime.tv_sec;
104 msg = readmsg(TSP_ANY, ANYADDR, &wait, 0);
106 (void)gettimeofday(&ntime, NULL);
107 if (ntime.tv_sec >= pollingtime) {
108 pollingtime = ntime.tv_sec + SAMPLEINTVL;
111 /* If a bogus master told us to quit, we can have decided to ignore a
112 * network. Therefore, periodically try to take over everything.
114 polls = (polls + 1) % POLLRATE;
115 if (0 == polls && nignorednets > 0) {
116 trace_msg("Looking for nets to re-master\n");
117 for (ntp = nettab; ntp; ntp = ntp->next) {
118 if (ntp->status == IGNORE
119 || ntp->status == NOMASTER) {
121 if (ntp->status == MASTER) {
126 if (ntp->status == MASTER
127 && --ntp->quit_count < 0)
136 for (ntp = nettab; ntp != NULL; ntp = ntp->next) {
137 to.tsp_type = TSP_LOOP;
138 to.tsp_vers = TSPVERSION;
139 to.tsp_seq = sequence++;
140 to.tsp_hopcnt = MAX_HOPCNT;
141 (void)strcpy(to.tsp_name, hostname);
143 if (sendto(sock, (char *)&to,
144 sizeof(struct tsp), 0,
145 (struct sockaddr*)&ntp->dest_addr,
146 sizeof(ntp->dest_addr)) < 0) {
147 trace_sendto_err(ntp->dest_addr.sin_addr);
154 switch (msg->tsp_type) {
165 * XXX check to see it is from ourself
167 tsp_time_sec = msg->tsp_time.tv_sec;
168 (void)strcpy(newdate, ctime(&tsp_time_sec));
169 if (!good_host_name(msg->tsp_name)) {
171 "attempted date change by %s to %s",
172 msg->tsp_name, newdate);
178 (void)gettimeofday(&ntime, NULL);
179 pollingtime = ntime.tv_sec + SAMPLEINTVL;
183 if (!fromnet || fromnet->status != MASTER)
185 tsp_time_sec = msg->tsp_time.tv_sec;
186 (void)strcpy(newdate, ctime(&tsp_time_sec));
187 htp = findhost(msg->tsp_name);
190 "attempted SET DATEREQ by uncontrolled %s to %s",
191 msg->tsp_name, newdate);
194 if (htp->seq == msg->tsp_seq)
196 htp->seq = msg->tsp_seq;
199 "attempted SET DATEREQ by untrusted %s to %s",
200 msg->tsp_name, newdate);
206 (void)gettimeofday(&ntime, NULL);
207 pollingtime = ntime.tv_sec + SAMPLEINTVL;
211 xmit(TSP_ACK, msg->tsp_seq, &from);
222 traceoff("Tracing ended at %s\n");
228 if (fromnet->status == MASTER) {
230 (void)addmach(msg->tsp_name, &from,fromnet);
233 (void)strcpy(tname, msg->tsp_name);
234 to.tsp_type = TSP_QUIT;
235 (void)strcpy(to.tsp_name, hostname);
236 answer = acksend(&to, &taddr, tname,
238 if (answer == NULL) {
239 syslog(LOG_ERR, "election error by %s",
246 * After a network partition, there can be
247 * more than one master: the first slave to
248 * come up will notify here the situation.
250 if (!fromnet || fromnet->status != MASTER)
252 (void)strcpy(to.tsp_name, hostname);
254 /* The other master often gets into the same state,
255 * with boring results if we stay at it forever.
257 ntp = fromnet; /* (acksend() can leave fromnet=0 */
258 for (i = 0; i < 3; i++) {
259 to.tsp_type = TSP_RESOLVE;
260 (void)strcpy(to.tsp_name, hostname);
261 answer = acksend(&to, &ntp->dest_addr,
262 ANYADDR, TSP_MASTERACK,
266 htp = addmach(answer->tsp_name,&from,ntp);
267 to.tsp_type = TSP_QUIT;
268 msg = acksend(&to, &htp->addr, htp->name,
269 TSP_ACK, 0, htp->noanswer);
272 "no response from %s to CONFLICT-QUIT",
281 if (!fromnet || fromnet->status != MASTER)
284 * do not want to call synch() while waiting
287 (void)gettimeofday(&ntime, NULL);
288 pollingtime = ntime.tv_sec + SAMPLEINTVL;
292 doquit(msg); /* become a slave */
296 if (!fromnet || fromnet->status != MASTER
297 || !strcmp(msg->tsp_name, hostname))
300 * We should not have received this from a net
301 * we are master on. There must be two masters.
303 htp = addmach(msg->tsp_name, &from,fromnet);
304 to.tsp_type = TSP_QUIT;
305 (void)strcpy(to.tsp_name, hostname);
306 answer = acksend(&to, &htp->addr, htp->name,
310 "loop breakage: no reply from %s=%s to QUIT",
311 htp->name, inet_ntoa(htp->addr.sin_addr));
318 "\tnets = %d, masters = %d, slaves = %d, ignored = %d\n",
319 nnets, nmasternets, nslavenets, nignorednets);
328 fprintf(fd, "garbage message: ");
339 * change the system date on the master
342 mchgdate(struct tsp *msg)
344 char tname[MAXHOSTNAMELEN];
346 struct timeval otime, ntime, tmptv;
349 (void)strcpy(tname, msg->tsp_name);
351 xmit(TSP_DATEACK, msg->tsp_seq, &from);
353 (void)strcpy(olddate, date());
355 /* adjust time for residence on the queue */
356 (void)gettimeofday(&otime, NULL);
357 adj_msg_time(msg,&otime);
359 tmptv.tv_sec = msg->tsp_time.tv_sec;
360 tmptv.tv_usec = msg->tsp_time.tv_usec;
361 timevalsub(&ntime, &tmptv, &otime);
362 if (ntime.tv_sec < MAXADJ && ntime.tv_sec > -MAXADJ) {
364 * do not change the clock if we can adjust it
367 synch(tvtomsround(ntime));
369 utx.ut_type = OLD_TIME;
370 (void)gettimeofday(&utx.ut_tv, NULL);
372 (void)settimeofday(&tmptv, 0);
373 utx.ut_type = NEW_TIME;
374 (void)gettimeofday(&utx.ut_tv, NULL);
379 syslog(LOG_NOTICE, "date changed by %s from %s",
385 * synchronize all of the slaves
392 struct timeval check, stop, wait;
396 fprintf(fd, "measurements starting at %s\n", date());
397 (void)gettimeofday(&check, NULL);
398 for (htp = self.l_fwd; htp != &self; htp = htp->l_fwd) {
399 if (htp->noanswer != 0) {
400 measure_status = measure(500, 100,
404 measure_status = measure(3000, 100,
408 if (measure_status != GOOD) {
409 /* The slave did not respond. We have
410 * just wasted lots of time on it.
412 htp->delta = HOSTDOWN;
413 if (++htp->noanswer >= LOSTHOST) {
416 "purging %s for not answering ICMP\n",
423 htp->delta = measure_delta;
425 (void)gettimeofday(&stop, NULL);
426 timevalsub(&stop, &stop, &check);
427 if (stop.tv_sec >= 1) {
431 * ack messages periodically
435 if (0 != readmsg(TSP_TRACEON,ANYADDR,
438 (void)gettimeofday(&check, NULL);
442 fprintf(fd, "measurements finished at %s\n", date());
444 if (!(status & SLAVE)) {
446 mydelta = networkdelta();
451 if (trace && (mydelta != 0 || (status & SLAVE)))
452 fprintf(fd,"local correction of %ld ms.\n", mydelta);
457 * sends the time to each slave after the master
458 * has received the command to set the network time
466 struct timeval tmptv;
468 /* Do not listen to the consensus after forcing the time. This is because
469 * the consensus takes a while to reach the time we are dictating.
472 for (htp = self.l_fwd; htp != &self; htp = htp->l_fwd) {
473 to.tsp_type = TSP_SETTIME;
474 (void)strcpy(to.tsp_name, hostname);
475 (void)gettimeofday(&tmptv, NULL);
476 to.tsp_time.tv_sec = tmptv.tv_sec;
477 to.tsp_time.tv_usec = tmptv.tv_usec;
478 answer = acksend(&to, &htp->addr, htp->name,
479 TSP_ACK, 0, htp->noanswer);
481 /* We client does not respond, then we have
482 * just wasted lots of time on it.
485 "no reply to SETTIME from %s", htp->name);
486 if (++htp->noanswer >= LOSTHOST) {
489 "purging %s for not answering",
502 static time_t next_time;
509 if (!fd) /* quit if tracing already off */
512 this_time = times(&tm);
513 if ((time_t)(this_time + delta) < next_time)
515 next_time = this_time + CLK_TCK;
517 fprintf(fd, "host table: %d entries at %s\n", slvcount, date());
520 for (i = 1; i <= slvcount; i++, htp = htp->l_fwd) {
521 l = strlen(htp->name) + 1;
522 if (length+l >= 80) {
527 fprintf(fd, " %s", htp->name);
533 static struct hosttbl *newhost_hash;
534 static struct hosttbl *lasthfree = &hosttbl[0];
537 struct hosttbl * /* answer or 0 */
545 for (p = name, i = 0; i < 8 && *p != '\0'; i++, p++)
547 newhost_hash = &hosttbl[j % NHOSTS];
550 if (htp->name[0] == '\0')
553 if (!strcmp(name, htp->name))
556 } while (htp != newhost_hash);
561 * add a host to the list of controlled machines if not already there
564 addmach(char *name, struct sockaddr_in *addr, struct netinfo *ntp)
566 struct hosttbl *ret, *p, *b, *f;
568 ret = findhost(name);
570 if (slvcount >= NHOSTS) {
572 fprintf(fd, "no more slots in host table\n");
575 syslog(LOG_ERR, "no more slots in host table");
577 longjmp(jmpenv, 2); /* give up and be a slave */
580 /* if our home hash slot is occupied, find a free entry
583 if (newhost_hash->name[0] != '\0') {
586 if (++lasthfree > &hosttbl[NHOSTS])
587 lasthfree = &hosttbl[1];
588 } while (ret->name[0] != '\0');
590 if (!newhost_hash->head) {
591 /* Move an interloper using our home. Use
592 * scratch pointers in case the new head is
593 * pointing to itself.
595 f = newhost_hash->h_fwd;
596 b = newhost_hash->h_bak;
599 f = newhost_hash->l_fwd;
600 b = newhost_hash->l_bak;
603 bcopy(newhost_hash,ret,sizeof(*ret));
609 /* link to an existing chain in our home
612 p = newhost_hash->h_bak;
613 ret->h_fwd = newhost_hash;
616 newhost_hash->h_bak = ret;
626 (void)strncpy(ret->name, name, sizeof(ret->name));
627 ret->good = good_host_name(name);
629 ret->l_bak = self.l_bak;
630 self.l_bak->l_fwd = ret;
638 ret->noanswer = (ret->noanswer != 0);
641 /* need to clear sequence number anyhow */
647 * remove the machine with the given index in the host table.
650 remmach(struct hosttbl *htp)
652 struct hosttbl *lprv, *hnxt, *f, *b;
655 fprintf(fd, "remove %s\n", htp->name);
657 /* get out of the lists */
658 htp->l_fwd->l_bak = lprv = htp->l_bak;
659 htp->l_bak->l_fwd = htp->l_fwd;
660 htp->h_fwd->h_bak = htp->h_bak;
661 htp->h_bak->h_fwd = hnxt = htp->h_fwd;
663 /* If we are in the home slot, pull up the chain */
664 if (htp->head && hnxt != htp) {
668 /* Use scratch pointers in case the new head is pointing to
680 bcopy(hnxt, htp, sizeof(*htp));
686 lasthfree->name[0] = '\0';
687 lasthfree->h_fwd = 0;
688 lasthfree->l_fwd = 0;
696 * Remove all the machines from the host table that exist on the given
697 * network. This is called when a master transitions to a slave on a
701 rmnetmachs(struct netinfo *ntp)
707 for (htp = self.l_fwd; htp != &self; htp = htp->l_fwd) {
716 masterup(struct netinfo *net)
718 xmit(TSP_MASTERUP, 0, &net->dest_addr);
721 * Do not tell new slaves our time for a while. This ensures
722 * we do not tell them to start using our time, before we have
723 * found a good master.
725 (void)gettimeofday(&net->slvwait, NULL);
729 newslave(struct tsp *msg)
732 struct tsp *answer, to;
733 struct timeval now, tmptv;
735 if (!fromnet || fromnet->status != MASTER)
738 htp = addmach(msg->tsp_name, &from,fromnet);
739 htp->seq = msg->tsp_seq;
744 * If we are stable, send our time to the slave.
745 * Do not go crazy if the date has been changed.
747 (void)gettimeofday(&now, NULL);
748 if (now.tv_sec >= fromnet->slvwait.tv_sec+3
749 || now.tv_sec < fromnet->slvwait.tv_sec) {
750 to.tsp_type = TSP_SETTIME;
751 (void)strcpy(to.tsp_name, hostname);
752 (void)gettimeofday(&tmptv, NULL);
753 to.tsp_time.tv_sec = tmptv.tv_sec;
754 to.tsp_time.tv_usec = tmptv.tv_usec;
755 answer = acksend(&to, &htp->addr,
762 "no reply to initial SETTIME from %s",
764 htp->noanswer = LOSTHOST;
771 * react to a TSP_QUIT:
774 doquit(struct tsp *msg)
776 if (fromnet->status == MASTER) {
777 if (!good_host_name(msg->tsp_name)) {
778 if (fromnet->quit_count <= 0) {
779 syslog(LOG_NOTICE,"untrusted %s told us QUIT",
781 suppress(&from, msg->tsp_name, fromnet);
782 fromnet->quit_count = 1;
785 syslog(LOG_NOTICE, "untrusted %s told us QUIT twice",
787 fromnet->quit_count = 2;
788 fromnet->status = NOMASTER;
790 fromnet->status = SLAVE;
793 longjmp(jmpenv, 2); /* give up and be a slave */
796 if (!good_host_name(msg->tsp_name)) {
797 syslog(LOG_NOTICE, "untrusted %s told us QUIT",
799 fromnet->quit_count = 2;
808 fd = fopen(_PATH_TIMEDLOG, "w");
813 fprintf(fd,"Tracing started at %s\n", date());
829 fprintf(fd, msg, date());