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[] = "$Id$";
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));
95 static char **get_server_list __P((char *file));
96 static char **get_single_server __P((void));
97 static void free_server_list __P((char **list));
99 static Unacked *srv_nacktab[SRV_NACKTAB_HASHSIZE];
100 Server *otherservers; /* points to an array of the known
102 int nservers; /* number of other servers */
103 int me_server_idx; /* # of my entry in the array */
105 #define ADJUST (1) /* adjust timeout on hello input */
106 #define DONT_ADJUST (0) /* don't adjust timeout */
108 /* parameters controlling the transitions of the FSM's--patchable with adb */
109 long timo_up = TIMO_UP;
110 long timo_tardy = TIMO_TARDY;
111 long timo_dead = TIMO_DEAD;
113 /* counters to measure old protocol use */
115 int old_compat_count_uloc = 0;
116 int old_compat_count_ulocate = 0;
117 int old_compat_count_subscr = 0;
118 #endif /* OLD_COMPAT */
120 int new_compat_count_uloc = 0;
121 int new_compat_count_subscr = 0;
122 #endif /* NEW_COMPAT */
128 * Initialize the array of servers. The `limbo' server goes in the first
129 * slot (otherservers[0]).
130 * Contact Hesiod to find all the other servers, allocate space for the
131 * structure, initialize them all to SERV_DEAD with expired timeouts.
132 * Set up a list header for server_forward retransmits.
139 struct in_addr *serv_addr, *server_addrs, limbo_addr;
141 /* we don't need to mask SIGFPE here since when we are called,
142 the signal handler isn't set up yet. */
144 /* talk to hesiod here, set nservers */
145 server_addrs = get_server_addrs(&nservers);
147 syslog(LOG_ERR, "No servers?!?");
156 /* increment servers to make room for 'limbo' */
159 otherservers = (Server *) malloc(nservers * sizeof(Server));
163 limbo_addr.s_addr = 0;
164 setup_server(otherservers, &limbo_addr);
165 timer_reset(otherservers[0].timer);
166 otherservers[0].timer = NULL;
167 otherservers[0].queue = NULL;
168 otherservers[0].dumping = 0;
170 for (serv_addr = server_addrs, i = 1; i < nservers; serv_addr++, i++) {
171 setup_server(&otherservers[i], serv_addr);
173 if (serv_addr->s_addr == my_addr.s_addr) {
175 otherservers[i].state = SERV_UP;
176 timer_reset(otherservers[i].timer);
177 otherservers[i].timer = NULL;
178 otherservers[i].queue = NULL;
179 otherservers[i].dumping = 0;
181 zdbug((LOG_DEBUG,"found myself"));
186 /* free up the addresses */
189 if (me_server_idx == -1) {
190 syslog(LOG_WARNING, "I'm a renegade server!");
191 otherservers = (Server *) realloc(otherservers,
192 ++nservers * sizeof(Server));
194 syslog(LOG_CRIT, "renegade realloc");
197 setup_server(&otherservers[nservers - 1], &my_addr);
199 otherservers[nservers - 1].state = SERV_UP;
201 /* I don't send hello's to myself--cancel the timer */
202 timer_reset(otherservers[nservers - 1].timer);
203 otherservers[nservers - 1].timer = NULL;
205 /* cancel and reschedule all the timers--pointers need
207 /* don't reschedule limbo's timer, so start i=1 */
208 for (i = 1; i < nservers - 1; i++) {
209 timer_reset(otherservers[i].timer);
210 /* all the HELLO's are due now */
211 otherservers[i].timer = timer_set_rel(0L, server_timo,
214 me_server_idx = nservers - 1;
220 * server_reset: re-initializes otherservers array by refreshing from Hesiod
223 * If any server is no longer named in the new list, and that server is in
224 * state SERV_DEAD, it is dropped from the server list.
225 * All other currently-known servers are retained.
226 * Any additional servers not previously known are added to the table.
228 * WARNING: Don't call this routine if any of the ancestor procedures have a
229 * handle on a particular server other than by indexing on otherservers[].
235 struct in_addr *server_addrs;
236 struct in_addr *serv_addr;
239 int *ok_list_new, *ok_list_old;
243 zdbug((LOG_DEBUG, "server_reset"));
247 syslog(LOG_INFO, "server_reset while alone, punt");
252 /* Find out what servers are supposed to be known. */
253 server_addrs = get_server_addrs(&num_servers);
255 syslog(LOG_ERR, "server_reset no servers. nothing done.");
258 ok_list_new = (int *) malloc(num_servers * sizeof(int));
260 syslog(LOG_ERR, "server_reset no mem new");
263 ok_list_old = (int *) malloc(nservers * sizeof(int));
265 syslog(LOG_ERR, "server_reset no mem old");
270 memset(ok_list_old, 0, nservers * sizeof(int));
271 memset(ok_list_new, 0, num_servers * sizeof(int));
273 /* reset timers--pointers will move */
274 for (j = 1; j < nservers; j++) { /* skip limbo */
275 if (j == me_server_idx)
277 timer_reset(otherservers[j].timer);
278 otherservers[j].timer = NULL;
281 /* check off entries on new list which are on old list.
282 check off entries on old list which are on new list. */
284 /* count limbo as "OK" */
286 ok_list_old[0] = 1; /* limbo is OK */
288 for (serv_addr = server_addrs, i = 0; i < num_servers; serv_addr++, i++) {
289 for (j = 1; j < nservers; j++) { /* j = 1 since we skip limbo */
290 if (otherservers[j].addr.sin_addr.s_addr == serv_addr->s_addr) {
291 /* if server is on both lists, mark */
295 break; /* for j loop */
300 /* remove any dead servers on old list not on new list. */
301 if (num_ok < nservers) {
304 new_num = 1; /* limbo */
305 /* count number of servers to keep */
306 for (j = 1; j < nservers; j++) {
307 /* since we are never SERV_DEAD, the following
308 test prevents removing ourself from the list */
309 if (ok_list_old[j] || (otherservers[j].state != SERV_DEAD)) {
310 syslog(LOG_INFO, "keeping server %s",
311 otherservers[j].addr_str);
315 if (new_num < nservers) {
316 servers = (Server *) malloc(new_num * sizeof(Server));
318 syslog(LOG_CRIT, "server_reset server malloc");
322 servers[0] = otherservers[0]; /* copy limbo */
324 srv = (int *) malloc(nservers * sizeof(int));
325 memset(srv, 0, nservers * sizeof(int));
327 /* copy the kept servers */
328 for (j = 1; j < nservers; j++) { /* skip limbo */
329 if (ok_list_old[j] ||
330 otherservers[j].state != SERV_DEAD) {
331 servers[i] = otherservers[j];
335 syslog(LOG_INFO, "flushing server %s",
336 otherservers[j].addr_str);
337 server_flush(&otherservers[j]);
342 srv_nack_renumber(srv);
346 otherservers = servers;
351 /* add any new servers on new list not on old list. */
353 for (i = 0; i < num_servers; i++) {
358 /* new_num is number of extras. */
360 otherservers = (Server *) realloc(otherservers, nservers * sizeof(Server));
362 syslog(LOG_CRIT, "server_reset realloc");
367 for (j = 1; j < nservers - new_num; j++) {
368 if (otherservers[j].addr.sin_addr.s_addr == my_addr.s_addr) {
373 if (!me_server_idx) {
374 syslog(LOG_CRIT, "can't find myself");
378 /* fill in otherservers with the new servers */
379 for (i = 0; i < num_servers; i++) {
380 if (!ok_list_new[i]) {
381 setup_server(&otherservers[nservers - (new_num--)],
383 syslog(LOG_INFO, "adding server %s", inet_ntoa(server_addrs[i]));
388 /* reset timers, to go off now.
389 We can't get a time-left indication (bleagh!)
390 so we expire them all now. This will generally
391 be non-destructive. We assume that when this code is
392 entered via a SIGHUP trigger that a system wizard
393 is watching the goings-on to make sure things straighten
396 for (i = 1; i < nservers; i++) { /* skip limbo */
397 if (i != me_server_idx && !otherservers[i].timer) {
398 otherservers[i].timer =
399 timer_set_rel(0L, server_timo, &otherservers[i]);
401 zdbug((LOG_DEBUG, "reset timer for %s",
402 otherservers[i].addr_str));
410 zdbug((LOG_DEBUG, "server_reset: %d servers now", nservers));
414 /* note: these must match the order given in zserver.h */
431 * A server timout has expired. If enough hello's have been unanswered,
432 * change state and act accordingly. Send a "hello" and reset the timer,
433 * incrementing the number of hello's sent.
435 * See the FSM in the Zephyr document for a better picture of what's
440 server_timo(void *arg)
442 Server *which = (Server *) arg;
446 zdbug((LOG_DEBUG,"srv_timo: %s", which->addr_str));
448 /* change state and reset if appropriate */
449 switch(which->state) {
450 case SERV_DEAD: /* leave him dead */
454 case SERV_UP: /* he's now tardy */
455 which->state = SERV_TARDY;
456 which->num_hello_sent = 0;
457 which->timeout = timo_tardy;
462 if (which->num_hello_sent >= ((which->state == SERV_TARDY) ?
465 /* he hasn't answered, assume DEAD */
466 which->state = SERV_DEAD;
467 which->num_hello_sent = 0;
468 which->timeout = timo_dead;
469 srv_nack_release(which);
474 syslog(LOG_ERR,"Bad server state, server 0x%x\n", (int)which);
477 /* now he's either TARDY, STARTING, or DEAD
478 We send a "hello," which increments the counter */
480 zdbug((LOG_DEBUG, "srv %s is %s", which->addr_str,
481 srv_states[which->state]));
483 server_hello(which, auth);
484 /* reschedule the timer */
485 which->timer = timer_set_rel(which->timeout, server_timo, which);
489 * Dispatch a notice from some other server
494 server_dispatch(ZNotice_t *notice,
496 struct sockaddr_in *who)
499 struct sockaddr_in newwho;
501 String *notice_class;
504 zdbug((LOG_DEBUG, "server_dispatch"));
507 if (notice->z_kind == SERVACK) {
508 srv_nack_cancel(notice, who);
512 /* set up a who for the real origin */
513 memset(&newwho, 0, sizeof(newwho));
514 newwho.sin_family = AF_INET;
515 newwho.sin_addr.s_addr = notice->z_sender_addr.s_addr;
516 newwho.sin_port = notice->z_port;
518 server = server_which_server(who);
520 /* we can dispatch to routines safely here, since they will
521 return ZSRV_REQUEUE if appropriate. We bounce this back
522 to the caller, and the caller will re-queue the message
523 for us to process later. */
525 notice_class = make_string(notice->z_class, 1);
527 if (realm_which_realm(&newwho))
528 status = realm_dispatch(notice, auth, &newwho, server);
529 else if (class_is_admin(notice_class)) {
530 /* admins don't get acked, else we get a packet loop */
531 /* will return requeue if bdump request and dumping */
533 return admin_dispatch(notice, auth, who, server);
534 } else if (class_is_control(notice_class)) {
535 status = control_dispatch(notice, auth, &newwho, server);
537 } else if (class_is_ulogin(notice_class)) {
538 status = ulogin_dispatch(notice, auth, &newwho, server);
540 } else if (class_is_ulocate(notice_class)) {
541 status = ulocate_dispatch(notice, auth, &newwho, server);
544 /* shouldn't come from another server */
545 syslog(LOG_WARNING, "srv_disp: pkt cls %s", notice->z_class);
546 status = ZERR_NONE; /* XXX */
548 if (status != ZSRV_REQUEUE)
549 ack(notice, who); /* acknowledge it if processed */
550 free_string(notice_class);
556 * Register a new server (one not in our list). This MUST be authenticated.
561 server_register(notice, auth, who)
564 struct sockaddr_in *who;
570 if (who->sin_port != srv_addr.sin_port) {
572 zdbug((LOG_DEBUG, "srv_wrong port %d", ntohs(who->sin_port)));
576 /* Not yet... talk to ken about authenticators */
580 zdbug((LOG_DEBUG, "srv_unauth"));
585 /* OK, go ahead and set him up. */
586 temp = (Server *) malloc((nservers + 1) * sizeof(Server));
588 syslog(LOG_CRIT, "srv_reg malloc");
592 memcpy(temp, otherservers, nservers * sizeof(Server));
595 /* don't reschedule limbo's timer, so start i=1 */
596 for (i = 1; i < nservers; i++) {
597 if (i == me_server_idx) /* don't reset myself */
599 /* reschedule the timers--we moved otherservers */
600 timerval = timer_when(otherservers[i].timer);
601 timer_reset(otherservers[i].timer);
602 otherservers[i].timer = timer_set_abs(timerval, server_timo,
605 setup_server(&otherservers[nservers], &who->sin_addr);
606 otherservers[nservers].state = SERV_STARTING;
607 otherservers[nservers].timeout = timo_tardy;
608 otherservers[nservers].update_queue = NULL;
609 otherservers[nservers].dumping = 0;
613 zdbug((LOG_DEBUG, "srv %s is %s", otherservers[nservers].addr_str,
614 srv_states[otherservers[nservers].state]));
622 * Tell the other servers that this client died.
626 server_kill_clt(Client *client)
629 char buf[512], *lyst[2];
631 ZNotice_t *pnotice; /* speed hack */
636 lyst[0] = inet_ntoa(client->addr.sin_addr),
637 sprintf(buf, "%d", ntohs(client->addr.sin_port));
641 zdbug((LOG_DEBUG, "server kill clt %s/%s", lyst[0], lyst[1]));
646 memset (¬ice, 0, sizeof(notice));
648 pnotice->z_kind = ACKED;
650 pnotice->z_port = srv_addr.sin_port;
651 pnotice->z_class = ZEPHYR_ADMIN_CLASS;
652 pnotice->z_class_inst = "";
653 pnotice->z_opcode = ADMIN_KILL_CLT;
654 pnotice->z_sender = myname; /* myname is the hostname */
655 pnotice->z_recipient = "";
656 pnotice->z_default_format = "";
657 pnotice->z_num_other_fields = 0;
662 /* don't tell limbo to flush, start at 1*/
663 for (i = 1; i < nservers; i++) {
664 if (i == me_server_idx) /* don't xmit to myself */
666 if (otherservers[i].state == SERV_DEAD)
669 retval = ZFormatNoticeList(pnotice, lyst, 2, &pack, &packlen,
670 auth ? ZAUTH : ZNOAUTH);
671 if (retval != ZERR_NONE) {
672 syslog(LOG_WARNING, "kill_clt format: %s", error_message(retval));
675 server_forw_reliable(&otherservers[i], pack, packlen, pnotice);
680 * A client has died. remove it
684 kill_clt(ZNotice_t *notice,
687 struct sockaddr_in who;
691 zdbug((LOG_DEBUG, "kill_clt"));
693 if (extract_addr(notice, &who) != ZERR_NONE)
694 return ZERR_NONE; /* XXX */
695 client = client_find(&who.sin_addr, notice->z_port);
697 syslog(LOG_NOTICE, "kill_clt: no such client (%s/%d) from %s",
698 inet_ntoa(who.sin_addr), ntohs(who.sin_port),
700 return ZERR_NONE; /* XXX */
704 syslog(LOG_DEBUG, "kill_clt clt_dereg %s/%d from %s",
705 inet_ntoa(who.sin_addr), ntohs(who.sin_port), server->addr_str);
709 /* remove the locations, too */
710 client_deregister(client, 1);
715 * extract a sockaddr_in from a message body
719 extract_addr(ZNotice_t *notice,
720 struct sockaddr_in *who)
722 char *cp = notice->z_message;
724 if (!notice->z_message_len) {
725 syslog(LOG_WARNING, "bad addr pkt");
728 who->sin_addr.s_addr = inet_addr(notice->z_message);
730 cp += strlen(cp) + 1;
731 if (cp >= notice->z_message + notice->z_message_len) {
732 syslog(LOG_WARNING, "short addr pkt");
735 who->sin_port = notice->z_port = htons((u_short) atoi(cp));
736 who->sin_family = AF_INET;
738 zdbug((LOG_DEBUG,"ext %s/%d", inet_ntoa(who->sin_addr),
739 ntohs(who->sin_port)));
745 * Flush all data associated with the server which
749 server_flush(Server *which)
753 syslog(LOG_DEBUG, "server_flush %s", which->addr_str);
755 srv_nack_release(which);
759 * send a hello to which, updating the count of hello's sent
760 * Authenticate if auth is set.
764 server_hello(Server *which,
767 send_msg(&which->addr, ADMIN_HELLO, auth);
768 which->num_hello_sent++;
772 * Handle an ADMIN message from a server
777 admin_dispatch(ZNotice_t *notice,
779 struct sockaddr_in *who,
782 char *opcode = notice->z_opcode;
783 Code_t status = ZERR_NONE;
786 zdbug((LOG_DEBUG, "ADMIN received"));
789 if (strcmp(opcode, ADMIN_HELLO) == 0) {
790 hello_respond(who, ADJUST, auth);
791 } else if (strcmp(opcode, ADMIN_IMHERE) == 0) {
793 } else if (strcmp(opcode, ADMIN_SHUTDOWN) == 0) {
795 zdbug((LOG_DEBUG, "server shutdown"));
798 srv_nack_release(server);
799 server->state = SERV_DEAD;
800 server->timeout = timo_dead;
801 /* don't worry about the timer, it will
802 be set appropriately on the next send */
804 zdbug((LOG_DEBUG, "srv %s is %s", server->addr_str,
805 srv_states[server->state]));
808 } else if (strcmp(opcode, ADMIN_BDUMP) == 0) {
809 /* Ignore a brain dump request if this is a brain dump packet
810 * or a packet being processed concurrently during a brain
812 if (bdumping || bdump_concurrent)
814 bdump_get(notice, auth, who, server);
815 } else if (strcmp(opcode, ADMIN_KILL_CLT) == 0) {
816 status = kill_clt(notice, server);
817 if (status == ZERR_NONE)
820 syslog(LOG_WARNING, "ADMIN unknown opcode %s",opcode);
827 * Handle an ADMIN message from some random client.
828 * For now, assume it's a registration-type message from some other
829 * previously unknown server
834 server_adispatch(ZNotice_t *notice,
836 struct sockaddr_in *who,
840 /* this had better be a HELLO message--start of acquisition
841 protocol, OR a status req packet */
843 if (strcmp(notice->z_opcode, ADMIN_STATUS) == 0) {
850 syslog(LOG_INFO, "disp: new server?");
851 if (server_register(notice, auth, who) != ZERR_NONE) {
852 syslog(LOG_INFO, "new server failed");
854 syslog(LOG_INFO, "new server %s, %d", inet_ntoa(who->sin_addr),
855 ntohs(who->sin_port));
856 hello_respond(who, DONT_ADJUST, auth);
859 syslog(LOG_INFO, "srv_adisp: server attempt from %s",
860 inet_ntoa(who->sin_addr));
867 send_stats(struct sockaddr_in *who)
873 char *vers, *pkts, *upt;
877 #define NUM_FIXED 3 /* 3 fixed fields, plus server info */
878 /* well, not really...but for
879 backward compatibility, we gotta
881 vers = get_version();
883 sprintf(buf, "%lu pkts", npackets);
885 sprintf(buf, "%ld seconds operational",NOW - uptime);
889 if (old_compat_count_uloc)
891 if (old_compat_count_ulocate)
893 if (old_compat_count_subscr)
895 #endif /* OLD_COMPAT */
897 if (new_compat_count_uloc)
899 if (new_compat_count_subscr)
901 #endif /* NEW_COMPAT */
902 extrafields += nrealms;
903 responses = (char **) malloc((NUM_FIXED + nservers + extrafields) *
909 num_resp = NUM_FIXED;
910 /* start at 1 and ignore limbo */
911 for (i = 1; i < nservers ; i++) {
912 sprintf(buf, "%s/%s%s", otherservers[i].addr_str,
913 srv_states[(int) otherservers[i].state],
914 otherservers[i].dumping ? " (DUMPING)" : "");
915 responses[num_resp++] = strsave(buf);
918 if (old_compat_count_uloc) {
919 sprintf(buf, "%d old old location requests", old_compat_count_uloc);
920 responses[num_resp++] = strsave(buf);
922 if (old_compat_count_ulocate) {
923 sprintf(buf, "%d old old loc lookup requests",
924 old_compat_count_ulocate);
925 responses[num_resp++] = strsave(buf);
927 if (old_compat_count_subscr) {
928 sprintf(buf, "%d old old subscr requests", old_compat_count_subscr);
929 responses[num_resp++] = strsave(buf);
931 #endif /* OLD_COMPAT */
933 if (new_compat_count_uloc) {
934 sprintf(buf, "%d new old location requests", new_compat_count_uloc);
935 responses[num_resp++] = strsave(buf);
937 if (new_compat_count_subscr) {
938 sprintf(buf, "%d new old subscr requests", new_compat_count_subscr);
939 responses[num_resp++] = strsave(buf);
941 #endif /* NEW_COMPAT */
942 for (realm = otherrealms, i = 0; i < nrealms ; i++, realm++) {
943 sprintf(buf, "%s(%s)/%s", realm->name,
944 inet_ntoa((realm->addrs[realm->idx]).sin_addr),
945 rlm_states[(int) realm->state]);
946 responses[num_resp++] = strsave(buf);
949 send_msg_list(who, ADMIN_STATUS, responses, num_resp, 0);
951 /* Start at one; don't try to free static version string */
952 for (i = 1; i < num_resp; i++)
958 * Get a list of server addresses.
960 * This list is retrieved from Hesiod.
962 * This list is read from a file.
964 * Return a pointer to an array of allocated storage. This storage is
965 * freed by the caller.
968 static struct in_addr *
969 get_server_addrs(int *number)
972 char **server_hosts = NULL;
973 char **server_hosts_free = NULL;
975 struct in_addr *addrs;
976 struct in_addr *addr;
979 server_hosts = get_server_list(list_file);
980 server_hosts_free = server_hosts;
983 server_hosts = hes_resolve("zephyr","sloc");
986 server_hosts = get_single_server();
987 server_hosts_free = server_hosts;
993 for (cpp = server_hosts; *cpp; cpp++)
996 addrs = (struct in_addr *) malloc(i * sizeof(struct in_addr));
998 /* Convert to in_addr's */
999 for (cpp = server_hosts, addr = addrs, i = 0; *cpp; cpp++) {
1000 hp = gethostbyname(*cpp);
1002 memcpy(addr, hp->h_addr, sizeof(struct in_addr));
1005 syslog(LOG_WARNING, "hostname failed, %s", *cpp);
1009 if (server_hosts_free)
1010 free_server_list(server_hosts_free);
1014 static int nhosts = 0;
1017 * read "file" to get a list of names of hosts to peer with.
1018 * The file should contain a list of host names, one per line.
1022 get_server_list(char *file)
1025 char buf[MAXHOSTNAMELEN];
1030 fp = fopen(file, "r");
1033 /* start with 16, realloc if necessary */
1035 ret_list = (char **) malloc(nhosts * sizeof(char *));
1039 while (fgets(buf, MAXHOSTNAMELEN, fp)) {
1040 /* nuke the newline, being careful not to overrun
1041 the buffer searching for it with strlen() */
1042 buf[MAXHOSTNAMELEN - 1] = '\0';
1043 newline = strchr(buf, '\n');
1047 if (nused + 1 >= nhosts) {
1048 /* get more pointer space if necessary */
1049 /* +1 to leave room for null pointer */
1050 ret_list = (char **) realloc(ret_list, nhosts * 2);
1051 nhosts = nhosts * 2;
1053 ret_list[nused++] = strsave(buf);
1060 ret_list[nused] = NULL;
1065 get_single_server(void)
1067 char buf[MAXHOSTNAMELEN];
1071 ret_list = (char **) malloc(nhosts * sizeof(char *));
1074 if (gethostname(buf, sizeof(buf)) < 0) {
1078 ret_list[nused++] = strsave(buf);
1079 ret_list[nused] = NULL;
1084 * free storage allocated by get_server_list
1087 free_server_list(char **list)
1089 char **orig_list = list;
1091 if (!nhosts) /* nothing allocated */
1093 for (; *list; list++)
1100 * initialize the server structure for address addr, and set a timer
1101 * to go off immediately to send hello's to other servers.
1105 setup_server(Server *server,
1106 struct in_addr *addr)
1108 server->state = SERV_DEAD;
1109 server->timeout = timo_dead;
1110 server->num_hello_sent = 0;
1111 server->addr.sin_family = AF_INET;
1112 /* he listens to the same port we do */
1113 server->addr.sin_port = srv_addr.sin_port;
1114 server->addr.sin_addr = *addr;
1115 strcpy(server->addr_str, inet_ntoa(*addr));
1116 server->timer = timer_set_rel(0L, server_timo, server);
1117 server->queue = NULL;
1118 server->dumping = 0;
1122 * Someone sent us a hello message, respond to them.
1126 hello_respond(struct sockaddr_in *who,
1133 zdbug((LOG_DEBUG, "hello from %s", inet_ntoa(who->sin_addr)));
1136 send_msg(who, ADMIN_IMHERE, auth);
1140 /* If we think he's down, schedule an immediate HELLO. */
1142 which = server_which_server(who);
1146 switch (which->state) {
1148 /* he said hello, we thought he was dead.
1149 reschedule his hello for now. */
1150 timer_reset(which->timer);
1151 which->timer = timer_set_rel(0L, server_timo, which);
1162 * return the server descriptor for server at who
1166 server_which_server(struct sockaddr_in *who)
1171 if (who->sin_port != srv_addr.sin_port)
1174 /* don't check limbo */
1175 for (server = &otherservers[1], i = 1; i < nservers; i++, server++) {
1176 if (server->addr.sin_addr.s_addr == who->sin_addr.s_addr)
1183 * We received a response to a hello packet or an ack. Adjust server state
1187 srv_responded(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.
1238 server_shutdown(void)
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(struct sockaddr_in *who,
1258 ZNotice_t *pnotice; /* speed hack */
1265 memset (¬ice, 0, sizeof(notice));
1267 pnotice->z_kind = ACKED;
1269 pnotice->z_port = srv_addr.sin_port;
1270 pnotice->z_class = ZEPHYR_ADMIN_CLASS;
1271 pnotice->z_class_inst = "";
1272 pnotice->z_opcode = opcode;
1273 pnotice->z_sender = myname; /* myname is the hostname */
1274 pnotice->z_recipient = "";
1275 pnotice->z_default_format = "";
1276 pnotice->z_message = NULL;
1277 pnotice->z_message_len = 0;
1278 pnotice->z_num_other_fields = 0;
1280 /* XXX for now, we don't do authentication */
1283 retval = ZFormatNotice(pnotice, &pack, &packlen, auth ? ZAUTH : ZNOAUTH);
1284 if (retval != ZERR_NONE) {
1285 syslog(LOG_WARNING, "snd_msg format: %s", error_message(retval));
1288 retval = ZSetDestAddr(who);
1289 if (retval != ZERR_NONE) {
1290 syslog(LOG_WARNING, "snd_msg set addr: %s", error_message(retval));
1294 /* don't wait for ack */
1295 retval = ZSendPacket(pack, packlen, 0);
1296 if (retval != ZERR_NONE)
1297 syslog(LOG_WARNING, "snd_msg xmit: %s", error_message(retval));
1302 * send a notice with a message to who with admin class and opcode and
1303 * message body as specified.
1304 * auth is set if we want to send authenticated
1305 * server_idx is -1 if we are sending to a client, or the server index
1306 * if we are sending to a server.
1310 send_msg_list(struct sockaddr_in *who,
1321 memset (¬ice, 0, sizeof(notice));
1323 notice.z_kind = UNSAFE;
1324 notice.z_port = srv_addr.sin_port;
1325 notice.z_class = ZEPHYR_ADMIN_CLASS;
1326 notice.z_class_inst = "";
1327 notice.z_opcode = opcode;
1328 notice.z_sender = myname; /* myname is the hostname */
1329 notice.z_recipient = "";
1330 notice.z_default_format = "";
1331 notice.z_message = NULL;
1332 notice.z_message_len = 0;
1333 notice.z_num_other_fields = 0;
1335 /* XXX for now, we don't do authentication */
1338 retval = ZFormatNoticeList(¬ice, lyst, num, &pack, &packlen,
1339 auth ? ZAUTH : ZNOAUTH);
1340 if (retval != ZERR_NONE) {
1341 syslog(LOG_WARNING, "snd_msg_lst format: %s", error_message(retval));
1344 retval = ZSetDestAddr(who);
1345 if (retval != ZERR_NONE) {
1346 syslog(LOG_WARNING, "snd_msg_lst set addr: %s", error_message(retval));
1350 xmit_frag(¬ice, pack, packlen, 0);
1355 * Forward the notice to the other servers
1359 server_forward(ZNotice_t *notice,
1361 struct sockaddr_in *who)
1369 zdbug((LOG_DEBUG, "srv_forw"));
1371 /* don't send to limbo */
1372 for (i = 1; i < nservers; i++) {
1373 if (i == me_server_idx) /* don't xmit to myself */
1375 if (otherservers[i].state == SERV_DEAD &&
1376 otherservers[i].dumping == 0) {
1377 /* if we are dumping to him, we want to
1378 queue it, even if he's dead */
1382 pack = malloc(sizeof(ZPacket_t));
1384 syslog(LOG_CRIT, "srv_fwd malloc");
1387 retval = ZNewFormatSmallRawNotice(notice, pack, &packlen);
1388 if (retval != ZERR_NONE) {
1389 syslog(LOG_WARNING, "srv_fwd format: %s", error_message(retval));
1392 if (otherservers[i].dumping) {
1393 server_queue(&otherservers[i], packlen, pack, auth, who);
1396 server_forw_reliable(&otherservers[i], pack, packlen, notice);
1401 server_forw_reliable(Server *server,
1410 retval = ZSetDestAddr(&server->addr);
1411 if (retval != ZERR_NONE) {
1412 syslog(LOG_WARNING, "srv_fwd_rel set addr: %s", error_message(retval));
1416 retval = ZSendPacket(pack, packlen, 0);
1417 if (retval != ZERR_NONE) {
1418 syslog(LOG_WARNING, "srv_fwd xmit: %s", error_message(retval));
1422 /* now we've sent it, mark it as not ack'ed */
1424 nacked = (Unacked *) malloc(sizeof(Unacked));
1426 /* no space: just punt */
1427 syslog(LOG_ERR, "srv_forw_rel nack malloc");
1432 nacked->client = NULL;
1433 nacked->rexmits = 0;
1434 nacked->packet = pack;
1435 nacked->dest.srv_idx = server - otherservers;
1436 nacked->packsz = packlen;
1437 nacked->uid = notice->z_uid;
1438 nacked->timer = timer_set_rel(rexmit_times[0], srv_rexmit, nacked);
1439 hashval = SRV_NACKTAB_HASHVAL(nacked->dest.srv_idx, nacked->uid);
1440 Unacked_insert(&srv_nacktab[hashval], nacked);
1444 * send the queued message for the server.
1448 server_send_queue(Server *server)
1454 while (server->queue) {
1455 pending = server_dequeue(server);
1456 status = ZParseNotice(pending->packet, pending->len, ¬ice);
1457 if (status != ZERR_NONE) {
1458 syslog(LOG_ERR, "ssq bad notice parse (%s): %s",
1459 inet_ntoa(pending->who.sin_addr), error_message(status));
1461 server_forw_reliable(server, pending->packet, pending->len,
1464 /* ACK handling routines will free the packet */
1470 * a server has acknowledged a message we sent to him; remove it from
1471 * server unacked queue
1475 srv_nack_cancel(ZNotice_t *notice,
1476 struct sockaddr_in *who)
1478 Server *server = server_which_server(who);
1483 syslog(LOG_ERR, "non-server ack?");
1486 hashval = SRV_NACKTAB_HASHVAL(server - otherservers, notice->z_uid);
1487 for (nacked = srv_nacktab[hashval]; nacked; nacked = nacked->next) {
1488 if (nacked->dest.srv_idx == server - otherservers
1489 && ZCompareUID(&nacked->uid, ¬ice->z_uid)) {
1490 timer_reset(nacked->timer);
1491 free(nacked->packet);
1492 Unacked_delete(nacked);
1498 zdbug((LOG_DEBUG, "srv_nack not found"));
1503 * retransmit a message to another server
1507 srv_rexmit(void *arg)
1509 Unacked *packet = (Unacked *) arg;
1511 /* retransmit the packet */
1514 zdbug((LOG_DEBUG,"srv_rexmit to %s/%d",
1515 otherservers[packet->dest.srv_idx].addr_str,
1516 ntohs(otherservers[packet->dest.srv_idx].addr.sin_port)));
1518 if (otherservers[packet->dest.srv_idx].state == SERV_DEAD) {
1520 zdbug((LOG_DEBUG, "cancelling send to dead server"));
1522 Unacked_delete(packet);
1523 free(packet->packet);
1524 srv_nack_release(&otherservers[packet->dest.srv_idx]);
1528 retval = ZSetDestAddr(&otherservers[packet->dest.srv_idx].addr);
1529 if (retval != ZERR_NONE) {
1530 syslog(LOG_WARNING, "srv_rexmit set addr: %s", error_message(retval));
1532 retval = ZSendPacket(packet->packet, packet->packsz, 0);
1533 if (retval != ZERR_NONE)
1534 syslog(LOG_WARNING, "srv_rexmit xmit: %s",
1535 error_message(retval));
1538 /* reset the timer */
1539 if (rexmit_times[packet->rexmits + 1] != -1)
1541 packet->timer = timer_set_rel(rexmit_times[packet->rexmits], srv_rexmit,
1546 * Clean up the not-yet-acked queue and release anything destined
1551 srv_nack_release(Server *server)
1554 Unacked *nacked, *next;
1556 for (i = 0; i < SRV_NACKTAB_HASHSIZE; i++) {
1557 for (nacked = srv_nacktab[i]; nacked; nacked = next) {
1558 next = nacked->next;
1559 if (nacked->dest.srv_idx == server - otherservers) {
1560 timer_reset(nacked->timer);
1561 Unacked_delete(nacked);
1562 free(nacked->packet);
1570 * Adjust indices of not-yet-acked packets sent to other servers to
1571 * continue to refer to the correct server.
1575 srv_nack_renumber (int *new_idx)
1577 /* XXX release any private queue for this server */
1581 /* search the not-yet-acked list for anything destined to 'from', and
1582 change the index to 'to'. */
1583 for (i = 0; i < SRV_NACKTAB_HASHSIZE; i++) {
1584 for (nacked = srv_nacktab[i]; nacked; nacked = nacked->next) {
1585 idx = new_idx[nacked->dest.srv_idx];
1587 syslog(LOG_ERR, "srv_nack_renumber error: [%d]=%d",
1588 nacked->dest.srv_idx, idx);
1591 nacked->dest.srv_idx = idx;
1597 * Queue this notice to be transmitted to the server when it is ready.
1600 server_queue(Server *server,
1604 struct sockaddr_in *who)
1608 pending = (Pending *) malloc(sizeof(Pending));
1610 syslog(LOG_CRIT, "update_queue malloc");
1613 pending->packet = pack;
1615 pending->auth = auth;
1616 pending->who = *who;
1617 pending->next = NULL;
1619 /* put it on the end of the list */
1621 server->queue_last->next = pending;
1623 server->queue = server->queue_last = pending;
1627 * Pull a notice off the hold queue.
1631 server_dequeue(Server *server)
1637 pending = server->queue;
1638 server->queue = pending->next;
1643 * free storage used by a pending queue entry.
1647 server_pending_free(Pending *pending)
1649 free(pending->packet);
1655 * Queue something to be handled later by this server.
1659 server_self_queue(ZNotice_t* notice,
1661 struct sockaddr_in * who)
1667 retval = ZFormatRawNotice(notice, &pack, &packlen);
1668 if (retval != ZERR_NONE) {
1669 syslog(LOG_CRIT, "srv_self_queue format: %s", error_message(retval));
1672 server_queue(me_server, packlen, pack, auth, who);
1676 * dump info about servers onto the fp.
1677 * assumed to be called with SIGFPE blocked
1678 * (true if called from signal handler)
1681 server_dump_servers(FILE *fp)
1685 for (i = 0; i < nservers ; i++) {
1686 fprintf(fp, "%d:%s/%s%s\n", i, otherservers[i].addr_str,
1687 srv_states[otherservers[i].state],
1688 otherservers[i].dumping ? " (DUMPING)" : "");