1 /* This file is part of the Project Athena Zephyr Notification System.
2 * It contains functions for communication with other servers.
4 * Created by: John T. Kohl
6 * $Source: /afs/dev.mit.edu/source/repository/athena/lib/zephyr/server/server.c,v $
9 * Copyright (c) 1987, 1991 by the Massachusetts Institute of Technology.
10 * For copying and distribution information, see the file
14 #include <zephyr/mit-copyright.h>
16 #include <sys/socket.h>
20 static const char rcsid_server_c[] = "$Id: server.c,v 1.66 2001/02/27 04:51:41 zacheiss Exp $";
24 #define SRV_NACKTAB_HASHSIZE 1023
25 #define SRV_NACKTAB_HASHVAL(which, uid) (((which) ^ (uid).zuid_addr.s_addr ^ \
26 (uid).tv.tv_sec ^ (uid).tv.tv_usec) \
27 % SRV_NACKTAB_HASHSIZE)
29 * Server manager. Deal with traffic to and from other servers.
33 * void server_shutdown()
35 * void server_timo(which)
38 * void server_dispatch(notice, auth, who)
41 * struct sockaddr_in *who;
43 * void server_recover(client)
46 * void server_adispatch(notice, auth, who, server)
49 * struct sockaddr_in *who;
52 * void server_forward(notice, auth, who)
55 * struct sockaddr_in *who;
57 * Server *server_which_server(who)
58 * struct sockaddr_in *who;
60 * void server_kill_clt(client);
63 * void server_dump_servers(fp);
66 * void server_reset();
69 static void server_flush __P((Server *));
70 static void hello_respond __P((struct sockaddr_in *, int, int));
71 static void srv_responded __P((struct sockaddr_in *));
72 static void send_msg __P((struct sockaddr_in *, char *, int));
73 static void send_msg_list __P((struct sockaddr_in *, char *, char **, int,
75 static void srv_nack_cancel __P((ZNotice_t *, struct sockaddr_in *));
76 static void srv_nack_release __P((Server *));
77 static void srv_nack_renumber __P((int *));
78 static void send_stats __P((struct sockaddr_in *));
79 static void server_queue __P((Server *, int, void *, int,
80 struct sockaddr_in *));
81 static void server_hello __P((Server *, int));
82 static void setup_server __P((Server *, struct in_addr *));
83 static void srv_rexmit __P((void *));
84 static void server_forw_reliable __P((Server *, caddr_t, int, ZNotice_t *));
85 static Code_t admin_dispatch __P((ZNotice_t *, int, struct sockaddr_in *,
87 static Code_t kill_clt __P((ZNotice_t *, Server *));
88 static Code_t extract_addr __P((ZNotice_t *, struct sockaddr_in *));
91 static Code_t server_register();
94 static struct in_addr *get_server_addrs __P((int *number));
96 static char **get_server_list __P((char *file));
97 static void free_server_list __P((char **list));
100 static Unacked *srv_nacktab[SRV_NACKTAB_HASHSIZE];
101 Server *otherservers; /* points to an array of the known
103 int nservers; /* number of other servers */
104 int me_server_idx; /* # of my entry in the array */
106 #define ADJUST (1) /* adjust timeout on hello input */
107 #define DONT_ADJUST (0) /* don't adjust timeout */
109 /* parameters controlling the transitions of the FSM's--patchable with adb */
110 long timo_up = TIMO_UP;
111 long timo_tardy = TIMO_TARDY;
112 long timo_dead = TIMO_DEAD;
114 /* counters to measure old protocol use */
116 int old_compat_count_uloc = 0;
117 int old_compat_count_ulocate = 0;
118 int old_compat_count_subscr = 0;
119 #endif /* OLD_COMPAT */
121 int new_compat_count_uloc = 0;
122 int new_compat_count_subscr = 0;
123 #endif /* NEW_COMPAT */
129 * Initialize the array of servers. The `limbo' server goes in the first
130 * slot (otherservers[0]).
131 * Contact Hesiod to find all the other servers, allocate space for the
132 * structure, initialize them all to SERV_DEAD with expired timeouts.
133 * Set up a list header for server_forward retransmits.
140 struct in_addr *serv_addr, *server_addrs, limbo_addr;
142 /* we don't need to mask SIGFPE here since when we are called,
143 the signal handler isn't set up yet. */
145 /* talk to hesiod here, set nservers */
146 server_addrs = get_server_addrs(&nservers);
148 syslog(LOG_ERR, "No servers?!?");
157 /* increment servers to make room for 'limbo' */
160 otherservers = (Server *) malloc(nservers * sizeof(Server));
164 limbo_addr.s_addr = 0;
165 setup_server(otherservers, &limbo_addr);
166 timer_reset(otherservers[0].timer);
167 otherservers[0].timer = NULL;
168 otherservers[0].queue = NULL;
169 otherservers[0].dumping = 0;
171 for (serv_addr = server_addrs, i = 1; i < nservers; serv_addr++, i++) {
172 setup_server(&otherservers[i], serv_addr);
174 if (serv_addr->s_addr == my_addr.s_addr) {
176 otherservers[i].state = SERV_UP;
177 timer_reset(otherservers[i].timer);
178 otherservers[i].timer = NULL;
179 otherservers[i].queue = NULL;
180 otherservers[i].dumping = 0;
182 zdbug((LOG_DEBUG,"found myself"));
187 /* free up the addresses */
190 if (me_server_idx == -1) {
191 syslog(LOG_WARNING, "I'm a renegade server!");
192 otherservers = (Server *) realloc(otherservers,
193 ++nservers * sizeof(Server));
195 syslog(LOG_CRIT, "renegade realloc");
198 setup_server(&otherservers[nservers - 1], &my_addr);
200 otherservers[nservers - 1].state = SERV_UP;
202 /* I don't send hello's to myself--cancel the timer */
203 timer_reset(otherservers[nservers - 1].timer);
204 otherservers[nservers - 1].timer = NULL;
206 /* cancel and reschedule all the timers--pointers need
208 /* don't reschedule limbo's timer, so start i=1 */
209 for (i = 1; i < nservers - 1; i++) {
210 timer_reset(otherservers[i].timer);
211 /* all the HELLO's are due now */
212 otherservers[i].timer = timer_set_rel(0L, server_timo,
215 me_server_idx = nservers - 1;
221 * server_reset: re-initializes otherservers array by refreshing from Hesiod
224 * If any server is no longer named in the new list, and that server is in
225 * state SERV_DEAD, it is dropped from the server list.
226 * All other currently-known servers are retained.
227 * Any additional servers not previously known are added to the table.
229 * WARNING: Don't call this routine if any of the ancestor procedures have a
230 * handle on a particular server other than by indexing on otherservers[].
236 struct in_addr *server_addrs;
237 struct in_addr *serv_addr;
240 int *ok_list_new, *ok_list_old;
244 zdbug((LOG_DEBUG, "server_reset"));
248 syslog(LOG_INFO, "server_reset while alone, punt");
253 /* Find out what servers are supposed to be known. */
254 server_addrs = get_server_addrs(&num_servers);
256 syslog(LOG_ERR, "server_reset no servers. nothing done.");
259 ok_list_new = (int *) malloc(num_servers * sizeof(int));
261 syslog(LOG_ERR, "server_reset no mem new");
264 ok_list_old = (int *) malloc(nservers * sizeof(int));
266 syslog(LOG_ERR, "server_reset no mem old");
271 memset(ok_list_old, 0, nservers * sizeof(int));
272 memset(ok_list_new, 0, num_servers * sizeof(int));
274 /* reset timers--pointers will move */
275 for (j = 1; j < nservers; j++) { /* skip limbo */
276 if (j == me_server_idx)
278 timer_reset(otherservers[j].timer);
279 otherservers[j].timer = NULL;
282 /* check off entries on new list which are on old list.
283 check off entries on old list which are on new list. */
285 /* count limbo as "OK" */
287 ok_list_old[0] = 1; /* limbo is OK */
289 for (serv_addr = server_addrs, i = 0; i < num_servers; serv_addr++, i++) {
290 for (j = 1; j < nservers; j++) { /* j = 1 since we skip limbo */
291 if (otherservers[j].addr.sin_addr.s_addr == serv_addr->s_addr) {
292 /* if server is on both lists, mark */
296 break; /* for j loop */
301 /* remove any dead servers on old list not on new list. */
302 if (num_ok < nservers) {
305 new_num = 1; /* limbo */
306 /* count number of servers to keep */
307 for (j = 1; j < nservers; j++) {
308 /* since we are never SERV_DEAD, the following
309 test prevents removing ourself from the list */
310 if (ok_list_old[j] || (otherservers[j].state != SERV_DEAD)) {
311 syslog(LOG_INFO, "keeping server %s",
312 otherservers[j].addr_str);
316 if (new_num < nservers) {
317 servers = (Server *) malloc(new_num * sizeof(Server));
319 syslog(LOG_CRIT, "server_reset server malloc");
323 servers[0] = otherservers[0]; /* copy limbo */
325 srv = (int *) malloc(nservers * sizeof(int));
326 memset(srv, 0, nservers * sizeof(int));
328 /* copy the kept servers */
329 for (j = 1; j < nservers; j++) { /* skip limbo */
330 if (ok_list_old[j] ||
331 otherservers[j].state != SERV_DEAD) {
332 servers[i] = otherservers[j];
336 syslog(LOG_INFO, "flushing server %s",
337 otherservers[j].addr_str);
338 server_flush(&otherservers[j]);
343 srv_nack_renumber(srv);
347 otherservers = servers;
352 /* add any new servers on new list not on old list. */
354 for (i = 0; i < num_servers; i++) {
359 /* new_num is number of extras. */
361 otherservers = (Server *) realloc(otherservers, nservers * sizeof(Server));
363 syslog(LOG_CRIT, "server_reset realloc");
368 for (j = 1; j < nservers - new_num; j++) {
369 if (otherservers[j].addr.sin_addr.s_addr == my_addr.s_addr) {
374 if (!me_server_idx) {
375 syslog(LOG_CRIT, "can't find myself");
379 /* fill in otherservers with the new servers */
380 for (i = 0; i < num_servers; i++) {
381 if (!ok_list_new[i]) {
382 setup_server(&otherservers[nservers - (new_num--)],
384 syslog(LOG_INFO, "adding server %s", inet_ntoa(server_addrs[i]));
389 /* reset timers, to go off now.
390 We can't get a time-left indication (bleagh!)
391 so we expire them all now. This will generally
392 be non-destructive. We assume that when this code is
393 entered via a SIGHUP trigger that a system wizard
394 is watching the goings-on to make sure things straighten
397 for (i = 1; i < nservers; i++) { /* skip limbo */
398 if (i != me_server_idx && !otherservers[i].timer) {
399 otherservers[i].timer =
400 timer_set_rel(0L, server_timo, &otherservers[i]);
402 zdbug((LOG_DEBUG, "reset timer for %s",
403 otherservers[i].addr_str));
411 zdbug((LOG_DEBUG, "server_reset: %d servers now", nservers));
415 /* note: these must match the order given in zserver.h */
432 * A server timout has expired. If enough hello's have been unanswered,
433 * change state and act accordingly. Send a "hello" and reset the timer,
434 * incrementing the number of hello's sent.
436 * See the FSM in the Zephyr document for a better picture of what's
444 Server *which = (Server *) arg;
448 zdbug((LOG_DEBUG,"srv_timo: %s", which->addr_str));
450 /* change state and reset if appropriate */
451 switch(which->state) {
452 case SERV_DEAD: /* leave him dead */
456 case SERV_UP: /* he's now tardy */
457 which->state = SERV_TARDY;
458 which->num_hello_sent = 0;
459 which->timeout = timo_tardy;
464 if (which->num_hello_sent >= ((which->state == SERV_TARDY) ?
467 /* he hasn't answered, assume DEAD */
468 which->state = SERV_DEAD;
469 which->num_hello_sent = 0;
470 which->timeout = timo_dead;
471 srv_nack_release(which);
476 syslog(LOG_ERR,"Bad server state, server 0x%x\n",which);
479 /* now he's either TARDY, STARTING, or DEAD
480 We send a "hello," which increments the counter */
482 zdbug((LOG_DEBUG, "srv %s is %s", which->addr_str,
483 srv_states[which->state]));
485 server_hello(which, auth);
486 /* reschedule the timer */
487 which->timer = timer_set_rel(which->timeout, server_timo, which);
491 * Dispatch a notice from some other server
496 server_dispatch(notice, auth, who)
499 struct sockaddr_in *who;
502 struct sockaddr_in newwho;
504 String *notice_class;
507 zdbug((LOG_DEBUG, "server_dispatch"));
510 if (notice->z_kind == SERVACK) {
511 srv_nack_cancel(notice, who);
515 /* set up a who for the real origin */
516 memset(&newwho, 0, sizeof(newwho));
517 newwho.sin_family = AF_INET;
518 newwho.sin_addr.s_addr = notice->z_sender_addr.s_addr;
519 newwho.sin_port = notice->z_port;
521 server = server_which_server(who);
523 /* we can dispatch to routines safely here, since they will
524 return ZSRV_REQUEUE if appropriate. We bounce this back
525 to the caller, and the caller will re-queue the message
526 for us to process later. */
528 notice_class = make_string(notice->z_class, 1);
530 if (realm_which_realm(&newwho))
531 status = realm_dispatch(notice, auth, &newwho, server);
532 else if (class_is_admin(notice_class)) {
533 /* admins don't get acked, else we get a packet loop */
534 /* will return requeue if bdump request and dumping */
536 return admin_dispatch(notice, auth, who, server);
537 } else if (class_is_control(notice_class)) {
538 status = control_dispatch(notice, auth, &newwho, server);
540 } else if (class_is_ulogin(notice_class)) {
541 status = ulogin_dispatch(notice, auth, &newwho, server);
543 } else if (class_is_ulocate(notice_class)) {
544 status = ulocate_dispatch(notice, auth, &newwho, server);
547 /* shouldn't come from another server */
548 syslog(LOG_WARNING, "srv_disp: pkt cls %s", notice->z_class);
549 status = ZERR_NONE; /* XXX */
551 if (status != ZSRV_REQUEUE)
552 ack(notice, who); /* acknowledge it if processed */
553 free_string(notice_class);
559 * Register a new server (one not in our list). This MUST be authenticated.
564 server_register(notice, auth, who)
567 struct sockaddr_in *who;
573 if (who->sin_port != srv_addr.sin_port) {
575 zdbug((LOG_DEBUG, "srv_wrong port %d", ntohs(who->sin_port)));
579 /* Not yet... talk to ken about authenticators */
583 zdbug((LOG_DEBUG, "srv_unauth"));
588 /* OK, go ahead and set him up. */
589 temp = (Server *) malloc((nservers + 1) * sizeof(Server));
591 syslog(LOG_CRIT, "srv_reg malloc");
595 memcpy(temp, otherservers, nservers * sizeof(Server));
598 /* don't reschedule limbo's timer, so start i=1 */
599 for (i = 1; i < nservers; i++) {
600 if (i == me_server_idx) /* don't reset myself */
602 /* reschedule the timers--we moved otherservers */
603 timerval = timer_when(otherservers[i].timer);
604 timer_reset(otherservers[i].timer);
605 otherservers[i].timer = timer_set_abs(timerval, server_timo,
608 setup_server(&otherservers[nservers], &who->sin_addr);
609 otherservers[nservers].state = SERV_STARTING;
610 otherservers[nservers].timeout = timo_tardy;
611 otherservers[nservers].update_queue = NULL;
612 otherservers[nservers].dumping = 0;
616 zdbug((LOG_DEBUG, "srv %s is %s", otherservers[nservers].addr_str,
617 srv_states[otherservers[nservers].state]));
625 * Tell the other servers that this client died.
629 server_kill_clt(client)
633 char buf[512], *lyst[2];
635 ZNotice_t *pnotice; /* speed hack */
640 lyst[0] = inet_ntoa(client->addr.sin_addr),
641 sprintf(buf, "%d", ntohs(client->addr.sin_port));
645 zdbug((LOG_DEBUG, "server kill clt %s/%s", lyst[0], lyst[1]));
650 memset (¬ice, 0, sizeof(notice));
652 pnotice->z_kind = ACKED;
654 pnotice->z_port = srv_addr.sin_port;
655 pnotice->z_class = ZEPHYR_ADMIN_CLASS;
656 pnotice->z_class_inst = "";
657 pnotice->z_opcode = ADMIN_KILL_CLT;
658 pnotice->z_sender = myname; /* myname is the hostname */
659 pnotice->z_recipient = "";
660 pnotice->z_default_format = "";
661 pnotice->z_num_other_fields = 0;
666 /* don't tell limbo to flush, start at 1*/
667 for (i = 1; i < nservers; i++) {
668 if (i == me_server_idx) /* don't xmit to myself */
670 if (otherservers[i].state == SERV_DEAD)
673 retval = ZFormatNoticeList(pnotice, lyst, 2, &pack, &packlen,
674 auth ? ZAUTH : ZNOAUTH);
675 if (retval != ZERR_NONE) {
676 syslog(LOG_WARNING, "kill_clt format: %s", error_message(retval));
679 server_forw_reliable(&otherservers[i], pack, packlen, pnotice);
684 * A client has died. remove it
688 kill_clt(notice, server)
692 struct sockaddr_in who;
696 zdbug((LOG_DEBUG, "kill_clt"));
698 if (extract_addr(notice, &who) != ZERR_NONE)
699 return ZERR_NONE; /* XXX */
700 client = client_find(&who.sin_addr, notice->z_port);
702 syslog(LOG_NOTICE, "kill_clt: no such client (%s/%d) from %s",
703 inet_ntoa(who.sin_addr), ntohs(who.sin_port),
705 return ZERR_NONE; /* XXX */
709 syslog(LOG_DEBUG, "kill_clt clt_dereg %s/%d from %s",
710 inet_ntoa(who.sin_addr), ntohs(who.sin_port), server->addr_str);
714 /* remove the locations, too */
715 client_deregister(client, 1);
720 * extract a sockaddr_in from a message body
724 extract_addr(notice, who)
726 struct sockaddr_in *who;
728 char *cp = notice->z_message;
730 if (!notice->z_message_len) {
731 syslog(LOG_WARNING, "bad addr pkt");
734 who->sin_addr.s_addr = inet_addr(notice->z_message);
736 cp += strlen(cp) + 1;
737 if (cp >= notice->z_message + notice->z_message_len) {
738 syslog(LOG_WARNING, "short addr pkt");
741 who->sin_port = notice->z_port = htons((u_short) atoi(cp));
742 who->sin_family = AF_INET;
744 zdbug((LOG_DEBUG,"ext %s/%d", inet_ntoa(who->sin_addr),
745 ntohs(who->sin_port)));
751 * Flush all data associated with the server which
760 syslog(LOG_DEBUG, "server_flush %s", which->addr_str);
762 srv_nack_release(which);
766 * send a hello to which, updating the count of hello's sent
767 * Authenticate if auth is set.
771 server_hello(which, auth)
775 send_msg(&which->addr, ADMIN_HELLO, auth);
776 which->num_hello_sent++;
780 * Handle an ADMIN message from a server
785 admin_dispatch(notice, auth, who, server)
788 struct sockaddr_in *who;
791 char *opcode = notice->z_opcode;
792 Code_t status = ZERR_NONE;
795 zdbug((LOG_DEBUG, "ADMIN received"));
798 if (strcmp(opcode, ADMIN_HELLO) == 0) {
799 hello_respond(who, ADJUST, auth);
800 } else if (strcmp(opcode, ADMIN_IMHERE) == 0) {
802 } else if (strcmp(opcode, ADMIN_SHUTDOWN) == 0) {
804 zdbug((LOG_DEBUG, "server shutdown"));
807 srv_nack_release(server);
808 server->state = SERV_DEAD;
809 server->timeout = timo_dead;
810 /* don't worry about the timer, it will
811 be set appropriately on the next send */
813 zdbug((LOG_DEBUG, "srv %s is %s", server->addr_str,
814 srv_states[server->state]));
817 } else if (strcmp(opcode, ADMIN_BDUMP) == 0) {
818 /* Ignore a brain dump request if this is a brain dump packet
819 * or a packet being processed concurrently during a brain
821 if (bdumping || bdump_concurrent)
823 bdump_get(notice, auth, who, server);
824 } else if (strcmp(opcode, ADMIN_KILL_CLT) == 0) {
825 status = kill_clt(notice, server);
826 if (status == ZERR_NONE)
829 syslog(LOG_WARNING, "ADMIN unknown opcode %s",opcode);
836 * Handle an ADMIN message from some random client.
837 * For now, assume it's a registration-type message from some other
838 * previously unknown server
843 server_adispatch(notice, auth, who, server)
846 struct sockaddr_in *who;
850 /* this had better be a HELLO message--start of acquisition
851 protocol, OR a status req packet */
853 if (strcmp(notice->z_opcode, ADMIN_STATUS) == 0) {
860 syslog(LOG_INFO, "disp: new server?");
861 if (server_register(notice, auth, who) != ZERR_NONE) {
862 syslog(LOG_INFO, "new server failed");
864 syslog(LOG_INFO, "new server %s, %d", inet_ntoa(who->sin_addr),
865 ntohs(who->sin_port));
866 hello_respond(who, DONT_ADJUST, auth);
869 syslog(LOG_INFO, "srv_adisp: server attempt from %s",
870 inet_ntoa(who->sin_addr));
878 struct sockaddr_in *who;
884 char *vers, *pkts, *upt;
888 #define NUM_FIXED 3 /* 3 fixed fields, plus server info */
889 /* well, not really...but for
890 backward compatibility, we gotta
892 vers = get_version();
894 sprintf(buf, "%d pkts", npackets);
896 sprintf(buf, "%d seconds operational",NOW - uptime);
900 if (old_compat_count_uloc)
902 if (old_compat_count_ulocate)
904 if (old_compat_count_subscr)
906 #endif /* OLD_COMPAT */
908 if (new_compat_count_uloc)
910 if (new_compat_count_subscr)
912 #endif /* NEW_COMPAT */
913 extrafields += nrealms;
914 responses = (char **) malloc((NUM_FIXED + nservers + extrafields) *
920 num_resp = NUM_FIXED;
921 /* start at 1 and ignore limbo */
922 for (i = 1; i < nservers ; i++) {
923 sprintf(buf, "%s/%s%s", otherservers[i].addr_str,
924 srv_states[(int) otherservers[i].state],
925 otherservers[i].dumping ? " (DUMPING)" : "");
926 responses[num_resp++] = strsave(buf);
929 if (old_compat_count_uloc) {
930 sprintf(buf, "%d old old location requests", old_compat_count_uloc);
931 responses[num_resp++] = strsave(buf);
933 if (old_compat_count_ulocate) {
934 sprintf(buf, "%d old old loc lookup requests",
935 old_compat_count_ulocate);
936 responses[num_resp++] = strsave(buf);
938 if (old_compat_count_subscr) {
939 sprintf(buf, "%d old old subscr requests", old_compat_count_subscr);
940 responses[num_resp++] = strsave(buf);
942 #endif /* OLD_COMPAT */
944 if (new_compat_count_uloc) {
945 sprintf(buf, "%d new old location requests", new_compat_count_uloc);
946 responses[num_resp++] = strsave(buf);
948 if (new_compat_count_subscr) {
949 sprintf(buf, "%d new old subscr requests", new_compat_count_subscr);
950 responses[num_resp++] = strsave(buf);
952 #endif /* NEW_COMPAT */
953 for (realm = otherrealms, i = 0; i < nrealms ; i++, realm++) {
954 sprintf(buf, "%s(%s)/%s", realm->name,
955 inet_ntoa((realm->addrs[realm->idx]).sin_addr),
956 rlm_states[(int) realm->state]);
957 responses[num_resp++] = strsave(buf);
960 send_msg_list(who, ADMIN_STATUS, responses, num_resp, 0);
962 /* Start at one; don't try to free static version string */
963 for (i = 1; i < num_resp; i++)
969 * Get a list of server addresses.
971 * This list is retrieved from Hesiod.
973 * This list is read from a file.
975 * Return a pointer to an array of allocated storage. This storage is
976 * freed by the caller.
979 static struct in_addr *
980 get_server_addrs(number)
981 int *number; /* RETURN */
986 struct in_addr *addrs;
987 struct in_addr *addr;
991 /* get the names from Hesiod */
992 server_hosts = hes_resolve("zephyr","sloc");
996 server_hosts = get_server_list(list_file);
1002 for (cpp = server_hosts; *cpp; cpp++)
1005 addrs = (struct in_addr *) malloc(i * sizeof(struct in_addr));
1007 /* Convert to in_addr's */
1008 for (cpp = server_hosts, addr = addrs, i = 0; *cpp; cpp++) {
1009 hp = gethostbyname(*cpp);
1011 memcpy(addr, hp->h_addr, sizeof(struct in_addr));
1014 syslog(LOG_WARNING, "hostname failed, %s", *cpp);
1019 free_server_list(server_hosts);
1026 static int nhosts = 0;
1029 * read "file" to get a list of names of hosts to peer with.
1030 * The file should contain a list of host names, one per line.
1034 get_server_list(file)
1038 char buf[MAXHOSTNAMELEN];
1043 /* start with 16, realloc if necessary */
1045 ret_list = (char **) malloc(nhosts * sizeof(char *));
1047 fp = fopen(file, "r");
1049 while (fgets(buf, MAXHOSTNAMELEN, fp)) {
1050 /* nuke the newline, being careful not to overrun
1051 the buffer searching for it with strlen() */
1052 buf[MAXHOSTNAMELEN - 1] = '\0';
1053 newline = strchr(buf, '\n');
1057 if (nused + 1 >= nhosts) {
1058 /* get more pointer space if necessary */
1059 /* +1 to leave room for null pointer */
1060 ret_list = (char **) realloc(ret_list, nhosts * 2);
1061 nhosts = nhosts * 2;
1063 ret_list[nused++] = strsave(buf);
1067 if (gethostname(buf, sizeof(buf)) < 0) {
1071 ret_list[nused++] = strsave(buf);
1073 ret_list[nused] = NULL;
1078 * free storage allocated by get_server_list
1081 free_server_list(list)
1084 char **orig_list = list;
1086 if (!nhosts) /* nothing allocated */
1088 for (; *list; list++)
1096 * initialize the server structure for address addr, and set a timer
1097 * to go off immediately to send hello's to other servers.
1101 setup_server(server, addr)
1103 struct in_addr *addr;
1105 server->state = SERV_DEAD;
1106 server->timeout = timo_dead;
1107 server->num_hello_sent = 0;
1108 server->addr.sin_family = AF_INET;
1109 /* he listens to the same port we do */
1110 server->addr.sin_port = srv_addr.sin_port;
1111 server->addr.sin_addr = *addr;
1112 strcpy(server->addr_str, inet_ntoa(*addr));
1113 server->timer = timer_set_rel(0L, server_timo, server);
1114 server->queue = NULL;
1115 server->dumping = 0;
1119 * Someone sent us a hello message, respond to them.
1123 hello_respond(who, adj, auth)
1124 struct sockaddr_in *who;
1131 zdbug((LOG_DEBUG, "hello from %s", inet_ntoa(who->sin_addr)));
1134 send_msg(who, ADMIN_IMHERE, auth);
1138 /* If we think he's down, schedule an immediate HELLO. */
1140 which = server_which_server(who);
1144 switch (which->state) {
1146 /* he said hello, we thought he was dead.
1147 reschedule his hello for now. */
1148 timer_reset(which->timer);
1149 which->timer = timer_set_rel(0L, server_timo, which);
1160 * return the server descriptor for server at who
1164 server_which_server(who)
1165 struct sockaddr_in *who;
1170 if (who->sin_port != srv_addr.sin_port)
1173 /* don't check limbo */
1174 for (server = &otherservers[1], i = 1; i < nservers; i++, server++) {
1175 if (server->addr.sin_addr.s_addr == who->sin_addr.s_addr)
1182 * We received a response to a hello packet or an ack. Adjust server state
1187 struct sockaddr_in *who;
1189 Server *which = server_which_server(who);
1192 zdbug((LOG_DEBUG, "srv_responded %s", inet_ntoa(who->sin_addr)));
1196 syslog(LOG_ERR, "hello input from non-server?!");
1200 switch (which->state) {
1202 /* he responded, we thought he was dead. mark as starting
1204 which->state = SERV_STARTING;
1205 which->timeout = timo_tardy;
1206 timer_reset(which->timer);
1207 which->timer = timer_set_rel(0L, server_timo, which);
1210 /* here we negotiate and set up a braindump */
1211 if (bdump_socket < 0)
1216 which->state = SERV_UP;
1220 /* reset the timer and counts */
1221 which->num_hello_sent = 0;
1222 which->timeout = timo_up;
1223 timer_reset(which->timer);
1224 which->timer = timer_set_rel(which->timeout, server_timo, which);
1228 zdbug((LOG_DEBUG, "srv %s is %s", which->addr_str,
1229 srv_states[which->state]));
1234 * Send each of the other servers a shutdown message.
1242 /* don't tell limbo to go away, start at 1*/
1243 for (i = 1; i < nservers; i++)
1244 send_msg(&otherservers[i].addr, ADMIN_SHUTDOWN, 1);
1248 * send a message to who with admin class and opcode and clinst as specified.
1249 * auth is set if we want to send authenticated
1253 send_msg(who, opcode, auth)
1254 struct sockaddr_in *who;
1259 ZNotice_t *pnotice; /* speed hack */
1266 memset (¬ice, 0, sizeof(notice));
1268 pnotice->z_kind = ACKED;
1270 pnotice->z_port = srv_addr.sin_port;
1271 pnotice->z_class = ZEPHYR_ADMIN_CLASS;
1272 pnotice->z_class_inst = "";
1273 pnotice->z_opcode = opcode;
1274 pnotice->z_sender = myname; /* myname is the hostname */
1275 pnotice->z_recipient = "";
1276 pnotice->z_default_format = "";
1277 pnotice->z_message = NULL;
1278 pnotice->z_message_len = 0;
1279 pnotice->z_num_other_fields = 0;
1281 /* XXX for now, we don't do authentication */
1284 retval = ZFormatNotice(pnotice, &pack, &packlen, auth ? ZAUTH : ZNOAUTH);
1285 if (retval != ZERR_NONE) {
1286 syslog(LOG_WARNING, "snd_msg format: %s", error_message(retval));
1289 retval = ZSetDestAddr(who);
1290 if (retval != ZERR_NONE) {
1291 syslog(LOG_WARNING, "snd_msg set addr: %s", error_message(retval));
1295 /* don't wait for ack */
1296 retval = ZSendPacket(pack, packlen, 0);
1297 if (retval != ZERR_NONE)
1298 syslog(LOG_WARNING, "snd_msg xmit: %s", error_message(retval));
1303 * send a notice with a message to who with admin class and opcode and
1304 * message body as specified.
1305 * auth is set if we want to send authenticated
1306 * server_idx is -1 if we are sending to a client, or the server index
1307 * if we are sending to a server.
1311 send_msg_list(who, opcode, lyst, num, auth)
1312 struct sockaddr_in *who;
1324 memset (¬ice, 0, sizeof(notice));
1326 notice.z_kind = UNSAFE;
1327 notice.z_port = srv_addr.sin_port;
1328 notice.z_class = ZEPHYR_ADMIN_CLASS;
1329 notice.z_class_inst = "";
1330 notice.z_opcode = opcode;
1331 notice.z_sender = myname; /* myname is the hostname */
1332 notice.z_recipient = "";
1333 notice.z_default_format = "";
1334 notice.z_message = NULL;
1335 notice.z_message_len = 0;
1336 notice.z_num_other_fields = 0;
1338 /* XXX for now, we don't do authentication */
1341 retval = ZFormatNoticeList(¬ice, lyst, num, &pack, &packlen,
1342 auth ? ZAUTH : ZNOAUTH);
1343 if (retval != ZERR_NONE) {
1344 syslog(LOG_WARNING, "snd_msg_lst format: %s", error_message(retval));
1347 retval = ZSetDestAddr(who);
1348 if (retval != ZERR_NONE) {
1349 syslog(LOG_WARNING, "snd_msg_lst set addr: %s", error_message(retval));
1353 xmit_frag(¬ice, pack, packlen, 0);
1358 * Forward the notice to the other servers
1362 server_forward(notice, auth, who)
1365 struct sockaddr_in *who;
1373 zdbug((LOG_DEBUG, "srv_forw"));
1375 /* don't send to limbo */
1376 for (i = 1; i < nservers; i++) {
1377 if (i == me_server_idx) /* don't xmit to myself */
1379 if (otherservers[i].state == SERV_DEAD &&
1380 otherservers[i].dumping == 0) {
1381 /* if we are dumping to him, we want to
1382 queue it, even if he's dead */
1386 pack = malloc(sizeof(ZPacket_t));
1388 syslog(LOG_CRIT, "srv_fwd malloc");
1391 retval = ZFormatSmallRawNotice(notice, pack, &packlen);
1392 if (retval != ZERR_NONE) {
1393 syslog(LOG_WARNING, "srv_fwd format: %s", error_message(retval));
1396 if (otherservers[i].dumping) {
1397 server_queue(&otherservers[i], packlen, pack, auth, who);
1400 server_forw_reliable(&otherservers[i], pack, packlen, notice);
1405 server_forw_reliable(server, pack, packlen, notice)
1415 retval = ZSetDestAddr(&server->addr);
1416 if (retval != ZERR_NONE) {
1417 syslog(LOG_WARNING, "srv_fwd_rel set addr: %s", error_message(retval));
1421 retval = ZSendPacket(pack, packlen, 0);
1422 if (retval != ZERR_NONE) {
1423 syslog(LOG_WARNING, "srv_fwd xmit: %s", error_message(retval));
1427 /* now we've sent it, mark it as not ack'ed */
1429 nacked = (Unacked *) malloc(sizeof(Unacked));
1431 /* no space: just punt */
1432 syslog(LOG_ERR, "srv_forw_rel nack malloc");
1437 nacked->client = NULL;
1438 nacked->rexmits = 0;
1439 nacked->packet = pack;
1440 nacked->dest.srv_idx = server - otherservers;
1441 nacked->packsz = packlen;
1442 nacked->uid = notice->z_uid;
1443 nacked->timer = timer_set_rel(rexmit_times[0], srv_rexmit, nacked);
1444 hashval = SRV_NACKTAB_HASHVAL(nacked->dest.srv_idx, nacked->uid);
1445 LIST_INSERT(&srv_nacktab[hashval], nacked);
1449 * send the queued message for the server.
1453 server_send_queue(server)
1460 while (server->queue) {
1461 pending = server_dequeue(server);
1462 status = ZParseNotice(pending->packet, pending->len, ¬ice);
1463 if (status != ZERR_NONE) {
1464 syslog(LOG_ERR, "ssq bad notice parse (%s): %s",
1465 inet_ntoa(pending->who.sin_addr), error_message(status));
1467 server_forw_reliable(server, pending->packet, pending->len,
1470 /* ACK handling routines will free the packet */
1476 * a server has acknowledged a message we sent to him; remove it from
1477 * server unacked queue
1481 srv_nack_cancel(notice, who)
1483 struct sockaddr_in *who;
1485 Server *server = server_which_server(who);
1490 syslog(LOG_ERR, "non-server ack?");
1493 hashval = SRV_NACKTAB_HASHVAL(server - otherservers, notice->z_uid);
1494 for (nacked = srv_nacktab[hashval]; nacked; nacked = nacked->next) {
1495 if (nacked->dest.srv_idx == server - otherservers
1496 && ZCompareUID(&nacked->uid, ¬ice->z_uid)) {
1497 timer_reset(nacked->timer);
1498 free(nacked->packet);
1499 LIST_DELETE(nacked);
1505 zdbug((LOG_DEBUG, "srv_nack not found"));
1510 * retransmit a message to another server
1517 Unacked *packet = (Unacked *) arg;
1519 /* retransmit the packet */
1522 zdbug((LOG_DEBUG,"srv_rexmit to %s/%d",
1523 otherservers[packet->dest.srv_idx].addr_str,
1524 ntohs(otherservers[packet->dest.srv_idx].addr.sin_port)));
1526 if (otherservers[packet->dest.srv_idx].state == SERV_DEAD) {
1528 zdbug((LOG_DEBUG, "cancelling send to dead server"));
1530 LIST_DELETE(packet);
1531 free(packet->packet);
1532 srv_nack_release(&otherservers[packet->dest.srv_idx]);
1536 retval = ZSetDestAddr(&otherservers[packet->dest.srv_idx].addr);
1537 if (retval != ZERR_NONE) {
1538 syslog(LOG_WARNING, "srv_rexmit set addr: %s", error_message(retval));
1540 retval = ZSendPacket(packet->packet, packet->packsz, 0);
1541 if (retval != ZERR_NONE)
1542 syslog(LOG_WARNING, "srv_rexmit xmit: %s",
1543 error_message(retval));
1546 /* reset the timer */
1547 if (rexmit_times[packet->rexmits + 1] != -1)
1549 packet->timer = timer_set_rel(rexmit_times[packet->rexmits], srv_rexmit,
1554 * Clean up the not-yet-acked queue and release anything destined
1559 srv_nack_release(server)
1563 Unacked *nacked, *next;
1565 for (i = 0; i < SRV_NACKTAB_HASHSIZE; i++) {
1566 for (nacked = srv_nacktab[i]; nacked; nacked = next) {
1567 next = nacked->next;
1568 if (nacked->dest.srv_idx == server - otherservers) {
1569 timer_reset(nacked->timer);
1570 LIST_DELETE(nacked);
1571 free(nacked->packet);
1579 * Adjust indices of not-yet-acked packets sent to other servers to
1580 * continue to refer to the correct server.
1584 srv_nack_renumber (new_idx)
1587 /* XXX release any private queue for this server */
1591 /* search the not-yet-acked list for anything destined to 'from', and
1592 change the index to 'to'. */
1593 for (i = 0; i < SRV_NACKTAB_HASHSIZE; i++) {
1594 for (nacked = srv_nacktab[i]; nacked; nacked = nacked->next) {
1595 idx = new_idx[nacked->dest.srv_idx];
1597 syslog(LOG_ERR, "srv_nack_renumber error: [%d]=%d",
1598 nacked->dest.srv_idx, idx);
1601 nacked->dest.srv_idx = idx;
1607 * Queue this notice to be transmitted to the server when it is ready.
1610 server_queue(server, len, pack, auth, who)
1615 struct sockaddr_in *who;
1619 pending = (Pending *) malloc(sizeof(Pending));
1621 syslog(LOG_CRIT, "update_queue malloc");
1624 pending->packet = pack;
1626 pending->auth = auth;
1627 pending->who = *who;
1628 pending->next = NULL;
1630 /* put it on the end of the list */
1632 server->queue_last->next = pending;
1634 server->queue = server->queue_last = pending;
1638 * Pull a notice off the hold queue.
1642 server_dequeue(server)
1649 pending = server->queue;
1650 server->queue = pending->next;
1655 * free storage used by a pending queue entry.
1659 server_pending_free(pending)
1662 free(pending->packet);
1668 * Queue something to be handled later by this server.
1672 server_self_queue(notice, auth, who)
1675 struct sockaddr_in * who;
1681 retval = ZFormatRawNotice(notice, &pack, &packlen);
1682 if (retval != ZERR_NONE) {
1683 syslog(LOG_CRIT, "srv_self_queue format: %s", error_message(retval));
1686 server_queue(me_server, packlen, pack, auth, who);
1690 * dump info about servers onto the fp.
1691 * assumed to be called with SIGFPE blocked
1692 * (true if called from signal handler)
1695 server_dump_servers(fp)
1700 for (i = 0; i < nservers ; i++) {
1701 fprintf(fp, "%d:%s/%s%s\n", i, otherservers[i].addr_str,
1702 srv_states[otherservers[i].state],
1703 otherservers[i].dumping ? " (DUMPING)" : "");