2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 #include <afsconfig.h>
11 #include <afs/param.h>
16 #include <sys/types.h>
23 #include <sys/socket.h>
24 #include <netinet/in.h>
32 #include <rx/rx_multi.h>
33 #include <afs/cellconfig.h>
35 #include <afs/afsutil.h>
36 #include <afs/netutils.h>
39 #define UBIK_INTERNALS
43 /* statics used to determine if we're the sync site */
44 static afs_int32 syncSiteUntil = 0; /* valid only if amSyncSite */
45 int ubik_amSyncSite = 0; /* flag telling if I'm sync site */
46 static nServers; /* total number of servers */
47 static char amIMagic = 0; /* is this host the magic host */
48 char amIClone = 0; /* is this a clone which doesn't vote */
49 static char ubik_singleServer = 0;
50 int (*ubik_CRXSecurityProc) ();
51 char *ubik_CRXSecurityRock;
52 afs_int32 ubikSecIndex;
53 struct rx_securityClass *ubikSecClass;
54 static verifyInterfaceAddress();
57 /* Module responsible for both deciding if we're currently the sync site,
58 * and keeping collecting votes so as to stay sync site.
60 * The basic module contacts all of the servers it can, trying to get them to vote
61 * for this server for sync site. The vote request message (called a beacon message)
62 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
63 * receiving sites to know how long the sync site guarantee is made for.
65 * Each of these beacon messages is thus both a declaration of how long this site will
66 * remain sync site, and an attempt to extend that time by collecting votes for a later
67 * sync site extension.
69 * The voting module is responsible for choosing a reasonable time until which it promises
70 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
71 * the interface (perhaps it should be?) but is instead a compile time constant that both
74 * The beacon and vote modules work intimately together; the vote module decides how long
75 * it should promise the beacon module its vote, and the beacon module takes all of these
76 * votes and decides for how long it is the synchronization site.
79 /* procedure called from debug rpc call to get this module's state for debugging */
81 register struct ubik_debug *aparm;
83 /* fill in beacon's state fields in the ubik_debug structure */
84 aparm->syncSiteUntil = syncSiteUntil;
85 aparm->nServers = nServers;
88 /* procedure that determines whether this site has enough current votes to remain sync site.
89 * called from higher-level modules (everything but the vote module).
91 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact
92 * light-weight process, haven't expired. We're sync site as long as a majority of the
93 * servers in existence have promised us unexpired guarantees. The variable ubik_syncSiteUntil
94 * contains the time at which the latest of the majority of the sync site guarantees expires
95 * (if the variable ubik_amSyncSite is true)
96 * This module also calls up to the recovery module if it thinks that the recovery module
97 * may have to pick up a new database (which offucr sif we lose the sync site votes).
101 register afs_int32 now;
102 register afs_int32 rcode;
104 /* special case for fast startup */
105 if (nServers == 1 && !amIClone) {
106 return 1; /* one guy is always the sync site */
109 if (ubik_amSyncSite == 0 || amIClone)
110 rcode = 0; /* if I don't think I'm the sync site, say so */
112 now = FT_ApproxTime();
113 if (syncSiteUntil <= now) { /* if my votes have expired, say so */
115 ubik_dprint("Ubik: I am no longer the sync site\n");
119 rcode = 1; /* otherwise still have the required votes */
123 urecovery_ResetState(); /* force recovery to re-execute */
124 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
128 /* setup server list; called with two parms, first is my address, second is list of other servers
129 * called only at initialization to set up the list of servers to contact for votes. Just creates
130 * the server structure. Note that there are two connections in every server structure, one for
131 * vote calls (which must always go through quickly) and one for database operations, which
132 * are subject to waiting for locks. If we used only one, the votes would sometimes get
133 * held up behind database operations, and the sync site guarantees would timeout
134 * even though the host would be up for communication.
136 * The "magic" host is the one with the lowest internet address. It is
137 * magic because its vote counts epsilon more than the others. This acts
138 * as a tie-breaker when we have an even number of hosts in the system.
139 * For example, if the "magic" host is up in a 2 site system, then it
140 * is sync site. Without the magic host hack, if anyone crashed in a 2
141 * site system, we'd be out of business.
143 ubeacon_InitServerListByInfo(ame, info, clones)
145 struct afsconf_cell *info;
150 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
154 ubeacon_InitServerList(ame, aservers)
156 register afs_int32 aservers[];
161 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
166 ubeacon_InitServerListCommon(ame, info, clones, aservers)
168 struct afsconf_cell *info;
170 register afs_int32 aservers[];
172 register struct ubik_server *ts;
174 register afs_int32 servAddr;
175 register afs_int32 i, code;
177 struct ubik_server *magicServer;
179 /* verify that the addresses passed in are correct */
180 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
183 /* get the security index to use, if we can */
184 if (ubik_CRXSecurityProc) {
185 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &ubikSecClass,
190 /* don't have sec module yet */
192 ubikSecClass = rxnull_NewClientSecurityObject();
194 magicHost = ntohl(ame); /* do comparisons in host order */
195 magicServer = (struct ubik_server *)0;
198 for (i = 0; i < info->numServers; i++) {
199 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
200 ntohl((afs_uint32) ame)) {
209 for (i = 0; i < info->numServers; i++) {
212 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
213 memset(ts, 0, sizeof(struct ubik_server));
214 ts->next = ubik_servers;
216 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
221 || ntohl((afs_uint32) ts->addr[0]) <
222 (afs_uint32) magicHost) {
223 magicHost = ntohl(ts->addr[0]);
230 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
231 ubik_callPortal, VOTE_SERVICE_ID,
232 ubikSecClass, ubikSecIndex);
235 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
236 ubik_callPortal, DISK_SERVICE_ID,
237 ubikSecClass, ubikSecIndex);
242 while ((servAddr = *aservers++)) {
244 return UNHOSTS; /* too many hosts */
245 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
246 memset(ts, 0, sizeof(struct ubik_server));
247 ts->next = ubik_servers;
249 ts->addr[0] = servAddr; /* primary address in net byte order */
250 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for vote reqs */
251 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for disk reqs */
252 ts->isClone = 0; /* don't know about clones */
254 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
255 magicHost = ntohl(servAddr);
262 magicServer->magic = 1; /* remember for when counting votes */
264 if (!amIClone && !magicServer)
268 ++nServers; /* count this server as well as the remotes */
270 nServers = i + 1; /* count this server as well as the remotes */
272 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
273 /* send addrs to all other servers */
274 code = updateUbikNetworkAddress(ubik_host);
278 /* Shoud we set some defaults for RX??
280 r_nRetries = (RPCTIMEOUT/r_retryInterval);
283 if (!ubik_servers) /* special case 1 server */
284 ubik_singleServer = 1;
285 if (nServers == 1 && !amIClone) {
286 ubik_amSyncSite = 1; /* let's start as sync site */
287 syncSiteUntil = 0x7fffffff; /* and be it quite a while */
290 if (nServers == 1) /* special case 1 server */
291 ubik_singleServer = 1;
294 if (ubik_singleServer) {
295 if (!ubik_amSyncSite)
296 ubik_dprint("Ubik: I am the sync site - 1 server\n");
298 syncSiteUntil = 0x7fffffff; /* quite a while */
303 /* main lwp loop for code that sends out beacons. This code only runs while
304 * we're sync site or we want to be the sync site. It runs in its very own light-weight
309 register afs_int32 code;
311 struct rx_connection *connections[MAXSERVERS];
312 struct ubik_server *servers[MAXSERVERS];
313 register afs_int32 i;
314 register struct ubik_server *ts;
315 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
316 struct ubik_tid ttid;
319 /* loop forever getting votes */
320 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
323 /* don't wakeup more than every POLLTIME seconds */
324 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
325 /* don't sleep if last collection phase took too long (probably timed someone out ) */
331 code = IOMGR_Select(0, 0, 0, 0, &tt);
335 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
337 if (ubik_singleServer)
338 continue; /* special-case 1 server for speedy startup */
340 if (!uvote_ShouldIRun())
341 continue; /* if voter has heard from a better candidate than us, don't bother running */
343 /* otherwise we should run for election, or we're the sync site (and have already won);
344 * send out the beacon packets */
345 /* build list of all up hosts (noticing dead hosts are running again
346 * is a task for the recovery module, not the beacon module), and
347 * prepare to send them an r multi-call containing the beacon message */
348 i = 0; /* collect connections */
349 for (ts = ubik_servers; ts; ts = ts->next) {
350 if (ts->up && ts->addr[0] != ubik_host[0]) {
352 connections[i++] = ts->vote_rxcid;
355 servers[i] = (struct ubik_server *)0; /* end of list */
356 /* note that we assume in the vote module that we'll always get at least BIGTIME
357 * seconds of vote from anyone who votes for us, which means we can conservatively
358 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
359 /* this next is essentially an expansion of rgen's ServBeacon routine */
361 ttid.epoch = ubik_epochTime;
362 if (ubik_dbase->flags & DBWRITING) {
364 * if a write is in progress, we have to send the writeTidCounter
365 * which holds the tid counter of the write transaction , and not
366 * send the tidCounter value which holds the tid counter of the
369 ttid.counter = ubik_dbase->writeTidCounter;
371 ttid.counter = ubik_dbase->tidCounter + 1;
372 #if defined(UBIK_PAUSE)
373 ubik_dbase->flags |= DBVOTING;
374 #endif /* UBIK_PAUSE */
376 /* now analyze return codes, counting up our votes */
377 yesVotes = 0; /* count how many to ensure we have quorum */
378 oldestYesVote = 0x7fffffff; /* time quorum expires */
379 syncsite = ubeacon_AmSyncSite();
380 startTime = FT_ApproxTime();
382 * Don't waste time using mult Rx calls if there are no connections out there
385 multi_Rx(connections, i) {
386 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
388 temp = FT_ApproxTime(); /* now, more or less */
389 ts = servers[multi_i];
390 ts->lastBeaconSent = temp;
392 /* note that the vote time (the return code) represents the time
393 * the vote was computed, *not* the time the vote expires. We compute
394 * the latter down below if we got enough votes to go with */
396 ts->lastVoteTime = code;
397 if (code < oldestYesVote)
398 oldestYesVote = code;
403 yesVotes++; /* the extra epsilon */
404 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
405 ts->beaconSinceDown = 1;
406 ubik_dprint("yes vote from host %s\n",
407 afs_inet_ntoa(ts->addr[0]));
408 } else if (code == 0) {
409 ts->lastVoteTime = temp;
411 ts->beaconSinceDown = 1;
412 ubik_dprint("no vote from %s\n",
413 afs_inet_ntoa(ts->addr[0]));
414 } else if (code < 0) {
416 ts->beaconSinceDown = 0;
417 urecovery_LostServer();
418 ubik_dprint("time out from %s\n",
419 afs_inet_ntoa(ts->addr[0]));
424 /* now call our own voter module to see if we'll vote for ourself. Note that
425 * the same restrictions apply for our voting for ourself as for our voting
426 * for anyone else. */
427 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
428 &ubik_dbase->version, &ttid);
432 yesVotes++; /* extra epsilon */
433 if (i < oldestYesVote)
436 #if defined(UBIK_PAUSE)
437 ubik_dbase->flags &= ~DBVOTING;
438 #endif /* UBIK_PAUSE */
440 /* now decide if we have enough votes to become sync site.
441 * Note that we can still get enough votes even if we didn't for ourself. */
442 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
443 if (!ubik_amSyncSite)
444 ubik_dprint("Ubik: I am the sync site\n");
446 syncSiteUntil = oldestYesVote + SMALLTIME;
447 LWP_NoYieldSignal(&ubik_amSyncSite);
450 ubik_dprint("Ubik: I am no longer the sync site\n");
452 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
459 * Input Param : ame is the pointer to my IP address specified in the
460 * CellServDB file. aservers is an array containing IP
461 * addresses of remote ubik servers. The array is
462 * terminated by a zero address.
464 * Algorithm : Verify that my IP addresses 'ame' does actually exist
465 * on this machine. If any of my IP addresses are there
466 * in the remote server list 'aserver', remove them from
467 * this list. Update global variable ubik_host[] with
470 * Return Values : 0 on success, non-zero on failure
473 verifyInterfaceAddress(ame, info, aservers)
474 afs_uint32 *ame; /* one of my interface addr in net byte order */
475 struct afsconf_cell *info;
476 afs_uint32 aservers[]; /* list of all possible server addresses */
478 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
479 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
480 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
484 totalServers = info->numServers;
485 else { /* count the number of servers */
486 for (totalServers = 0, servList = aservers; *servList; servList++)
491 /* get all my interface addresses in net byte order */
492 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
494 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
496 * Find addresses we are supposed to register as per the netrestrict file
497 * if it exists, else just register all the addresses we find on this
498 * host as returned by rx_getAllAddr (in NBO)
502 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
503 AFSDIR_SERVER_NETINFO_FILEPATH,
504 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
506 ubik_print("ubik: Can't register any valid addresses:%s\n",
508 ubik_print("Aborting..\n");
513 /* get all my interface addresses in net byte order */
514 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
518 if (count <= 0) { /* no address found */
519 ubik_print("ubik: No network addresses found, aborting..");
523 /* verify that the My-address passed in by ubik is correct */
524 for (j = 0, found = 0; j < count; j++) {
525 if (*ame == myAddr[j]) { /* both in net byte order */
532 ubik_print("ubik: primary address %s does not exist\n",
533 afs_inet_ntoa(*ame));
534 /* if we had the result of rx_getAllAddr already, avoid subverting
535 * the "is gethostbyname(gethostname()) us" check. If we're
536 * using NetInfo/NetRestrict, we assume they have enough clue
537 * to avoid that big hole in their foot from the loaded gun. */
539 /* take the address we did get, then see if ame was masked */
541 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
542 if (tcount <= 0) { /* no address found */
543 ubik_print("ubik: No network addresses found, aborting..");
547 /* verify that the My-address passed in by ubik is correct */
548 for (j = 0, found = 0; j < tcount; j++) {
549 if (*ame == myAddr2[j]) { /* both in net byte order */
559 /* if any of my addresses are there in serverList, then
560 ** use that as my primary addresses : the higher level
561 ** application screwed up in dealing with multihomed concepts
563 for (j = 0, found = 0; j < count; j++) {
564 for (i = 0; i < totalServers; i++) {
566 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
568 tmpAddr = aservers[i];
569 if (myAddr[j] == tmpAddr) {
578 ubik_print("Using %s as my primary address\n", afs_inet_ntoa(*ame));
581 /* get rid of servers which were purged because all
582 ** those interface addresses are myself
584 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
585 /* find the first zero entry from the beginning */
586 for (; (start < end) && (aservers[start]); start++);
588 /* find the last non-zero entry from the end */
589 for (; (end >= 0) && (!aservers[end]); end--);
591 /* if there is nothing more to purge, exit from loop */
596 aservers[start] = aservers[end];
597 aservers[end] = 0; /* this entry was moved */
601 /* update all my addresses in ubik_host in such a way
602 ** that ubik_host[0] has the primary address
605 for (j = 0, i = 1; j < count; j++)
606 if (*ame != myAddr[j])
607 ubik_host[i++] = myAddr[j];
609 return 0; /* return success */
614 * Input Param : ubik_host is an array containing all my IP addresses.
616 * Algorithm : Do an RPC to all remote ubik servers infroming them
617 * about my IP addresses. Get their IP addresses and
618 * update my linked list of ubik servers 'ubik_servers'
620 * Return Values : 0 on success, non-zero on failure
623 updateUbikNetworkAddress(ubik_host)
624 afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR];
626 int j, count, code = 0;
627 UbikInterfaceAddr inAddr, outAddr;
628 struct rx_connection *conns[MAXSERVERS];
629 struct ubik_server *ts, *server[MAXSERVERS];
632 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
633 conns[count] = ts->disk_rxcid;
638 /* inform all other servers only if there are more than one
639 * database servers in the cell */
643 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
644 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
647 /* do the multi-RX RPC to all other servers */
648 multi_Rx(conns, count) {
649 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
650 ts = server[multi_i]; /* reply received from this server */
652 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
654 strcpy(buffer, (char *)afs_inet_ntoa(ts->addr[0]));
655 ubik_print("ubik:Two primary addresses for same server \
656 %s %s\n", buffer, afs_inet_ntoa(htonl(outAddr.hostAddr[0])));
658 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
659 ts->addr[j] = htonl(outAddr.hostAddr[j]);
661 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
663 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
664 afs_inet_ntoa(ts->addr[0]));
665 } else if (multi_error == UBADHOST) {
666 code = UBADHOST; /* remote CellServDB inconsistency */
667 ubik_print("Inconsistent Cell Info on server: ");
668 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
669 ubik_print("%s ", afs_inet_ntoa(ts->addr[j]));
672 ts->up = 0; /* mark the remote server as down */