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>
15 #include <sys/types.h>
26 #include <sys/socket.h>
27 #include <netinet/in.h>
34 #include <rx/rx_multi.h>
35 #include <afs/cellconfig.h>
37 #include <afs/afsutil.h>
38 #include <afs/netutils.h>
41 #define UBIK_INTERNALS
45 /*! \name statics used to determine if we're the sync site */
46 static afs_int32 syncSiteUntil = 0; /*!< valid only if amSyncSite */
47 int ubik_amSyncSite = 0; /*!< flag telling if I'm sync site */
48 static int nServers; /*!< total number of servers */
49 static char amIMagic = 0; /*!< is this host the magic host */
50 char amIClone = 0; /*!< is this a clone which doesn't vote */
51 static char ubik_singleServer = 0;
53 int (*ubik_CRXSecurityProc) (void *rock, struct rx_securityClass **,
55 void *ubik_CRXSecurityRock;
56 afs_int32 ubikSecIndex;
57 struct rx_securityClass *ubikSecClass;
58 static int ubeacon_InitServerListCommon(afs_uint32 ame,
59 struct afsconf_cell *info,
61 afs_uint32 aservers[]);
62 static int verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
63 afs_uint32 aservers[]);
64 static int updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR]);
68 * Module responsible for both deciding if we're currently the sync site,
69 * and keeping collecting votes so as to stay sync site.
71 * The basic module contacts all of the servers it can, trying to get them to vote
72 * for this server for sync site. The vote request message (called a beacon message)
73 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
74 * receiving sites to know how long the sync site guarantee is made for.
76 * Each of these beacon messages is thus both a declaration of how long this site will
77 * remain sync site, and an attempt to extend that time by collecting votes for a later
78 * sync site extension.
80 * The voting module is responsible for choosing a reasonable time until which it promises
81 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
82 * the interface (perhaps it should be?) but is instead a compile time constant that both
85 * The beacon and vote modules work intimately together; the vote module decides how long
86 * it should promise the beacon module its vote, and the beacon module takes all of these
87 * votes and decides for how long it is the synchronization site.
90 /*! \brief procedure called from debug rpc call to get this module's state for debugging */
92 ubeacon_Debug(struct ubik_debug *aparm)
94 /* fill in beacon's state fields in the ubik_debug structure */
95 aparm->syncSiteUntil = syncSiteUntil;
96 aparm->nServers = nServers;
100 * \brief Procedure that determines whether this site has enough current votes to remain sync site.
102 * Called from higher-level modules (everything but the vote module).
104 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact()
105 * light-weight process, haven't expired. We're sync site as long as a majority of the
106 * servers in existence have promised us unexpired guarantees. The variable #ubik_syncSiteUntil
107 * contains the time at which the latest of the majority of the sync site guarantees expires
108 * (if the variable #ubik_amSyncSite is true)
109 * This module also calls up to the recovery module if it thinks that the recovery module
110 * may have to pick up a new database (which offucr sif [sic] we lose the sync site votes).
112 * \return 1 if local site is the sync site
113 * \return 0 if sync site is elsewhere
116 ubeacon_AmSyncSite(void)
121 /* special case for fast startup */
122 if (nServers == 1 && !amIClone) {
123 return 1; /* one guy is always the sync site */
126 if (ubik_amSyncSite == 0 || amIClone)
127 rcode = 0; /* if I don't think I'm the sync site, say so */
129 now = FT_ApproxTime();
130 if (syncSiteUntil <= now) { /* if my votes have expired, say so */
132 ubik_dprint("Ubik: I am no longer the sync site\n");
136 rcode = 1; /* otherwise still have the required votes */
140 urecovery_ResetState(); /* force recovery to re-execute */
141 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
146 * \see ubeacon_InitServerListCommon()
149 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
154 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
159 * \param ame "address of me"
160 * \param aservers list of other servers
162 * \see ubeacon_InitServerListCommon()
165 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
170 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
176 * \brief setup server list
178 * \param ame "address of me"
179 * \param aservers list of other servers
181 * called only at initialization to set up the list of servers to
182 * contact for votes. Just creates the server structure.
184 * The "magic" host is the one with the lowest internet address. It is
185 * magic because its vote counts epsilon more than the others. This acts
186 * as a tie-breaker when we have an even number of hosts in the system.
187 * For example, if the "magic" host is up in a 2 site system, then it
188 * is sync site. Without the magic host hack, if anyone crashed in a 2
189 * site system, we'd be out of business.
191 * \note There are two connections in every server structure, one for
192 * vote calls (which must always go through quickly) and one for database
193 * operations, which are subject to waiting for locks. If we used only
194 * one, the votes would sometimes get held up behind database operations,
195 * and the sync site guarantees would timeout even though the host would be
196 * up for communication.
198 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
201 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
202 char clones[], afs_uint32 aservers[])
204 struct ubik_server *ts;
209 struct ubik_server *magicServer;
211 /* verify that the addresses passed in are correct */
212 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
215 /* get the security index to use, if we can */
216 if (ubik_CRXSecurityProc) {
217 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &ubikSecClass,
222 /* don't have sec module yet */
224 ubikSecClass = rxnull_NewClientSecurityObject();
226 magicHost = ntohl(ame); /* do comparisons in host order */
227 magicServer = (struct ubik_server *)0;
230 for (i = 0; i < info->numServers; i++) {
231 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
232 ntohl((afs_uint32) ame)) {
241 for (i = 0; i < info->numServers; i++) {
244 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
245 memset(ts, 0, sizeof(struct ubik_server));
246 ts->next = ubik_servers;
248 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
253 || ntohl((afs_uint32) ts->addr[0]) <
254 (afs_uint32) magicHost) {
255 magicHost = ntohl(ts->addr[0]);
262 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
263 ubik_callPortal, VOTE_SERVICE_ID,
264 ubikSecClass, ubikSecIndex);
267 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
268 ubik_callPortal, DISK_SERVICE_ID,
269 ubikSecClass, ubikSecIndex);
274 while ((servAddr = *aservers++)) {
276 return UNHOSTS; /* too many hosts */
277 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
278 memset(ts, 0, sizeof(struct ubik_server));
279 ts->next = ubik_servers;
281 ts->addr[0] = servAddr; /* primary address in net byte order */
282 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for vote reqs */
283 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for disk reqs */
284 ts->isClone = 0; /* don't know about clones */
286 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
287 magicHost = ntohl(servAddr);
294 magicServer->magic = 1; /* remember for when counting votes */
296 if (!amIClone && !magicServer)
300 ++nServers; /* count this server as well as the remotes */
302 nServers = i + 1; /* count this server as well as the remotes */
304 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
305 /* send addrs to all other servers */
306 code = updateUbikNetworkAddress(ubik_host);
310 /* Shoud we set some defaults for RX??
312 r_nRetries = (RPCTIMEOUT/r_retryInterval);
315 if (!ubik_servers) /* special case 1 server */
316 ubik_singleServer = 1;
317 if (nServers == 1 && !amIClone) {
318 ubik_amSyncSite = 1; /* let's start as sync site */
319 syncSiteUntil = 0x7fffffff; /* and be it quite a while */
322 if (nServers == 1) /* special case 1 server */
323 ubik_singleServer = 1;
326 if (ubik_singleServer) {
327 if (!ubik_amSyncSite)
328 ubik_dprint("Ubik: I am the sync site - 1 server\n");
330 syncSiteUntil = 0x7fffffff; /* quite a while */
336 * \brief main lwp loop for code that sends out beacons.
338 * This code only runs while we're sync site or we want to be the sync site.
339 * It runs in its very own light-weight process.
342 ubeacon_Interact(void *dummy)
346 struct rx_connection *connections[MAXSERVERS];
347 struct ubik_server *servers[MAXSERVERS];
349 struct ubik_server *ts;
350 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
351 struct ubik_tid ttid;
354 /* loop forever getting votes */
355 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
358 /* don't wakeup more than every POLLTIME seconds */
359 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
360 /* don't sleep if last collection phase took too long (probably timed someone out ) */
366 #ifdef AFS_PTHREAD_ENV
367 code = select(0, 0, 0, 0, &tt);
369 code = IOMGR_Select(0, 0, 0, 0, &tt);
374 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
376 if (ubik_singleServer)
377 continue; /* special-case 1 server for speedy startup */
379 if (!uvote_ShouldIRun())
380 continue; /* if voter has heard from a better candidate than us, don't bother running */
382 /* otherwise we should run for election, or we're the sync site (and have already won);
383 * send out the beacon packets */
384 /* build list of all up hosts (noticing dead hosts are running again
385 * is a task for the recovery module, not the beacon module), and
386 * prepare to send them an r multi-call containing the beacon message */
387 i = 0; /* collect connections */
388 for (ts = ubik_servers; ts; ts = ts->next) {
389 if (ts->up && ts->addr[0] != ubik_host[0]) {
391 connections[i++] = ts->vote_rxcid;
394 servers[i] = (struct ubik_server *)0; /* end of list */
395 /* note that we assume in the vote module that we'll always get at least BIGTIME
396 * seconds of vote from anyone who votes for us, which means we can conservatively
397 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
398 /* this next is essentially an expansion of rgen's ServBeacon routine */
400 ttid.epoch = ubik_epochTime;
401 if (ubik_dbase->flags & DBWRITING) {
403 * if a write is in progress, we have to send the writeTidCounter
404 * which holds the tid counter of the write transaction , and not
405 * send the tidCounter value which holds the tid counter of the
408 ttid.counter = ubik_dbase->writeTidCounter;
410 ttid.counter = ubik_dbase->tidCounter + 1;
411 #if defined(UBIK_PAUSE)
412 ubik_dbase->flags |= DBVOTING;
413 #endif /* UBIK_PAUSE */
415 /* now analyze return codes, counting up our votes */
416 yesVotes = 0; /* count how many to ensure we have quorum */
417 oldestYesVote = 0x7fffffff; /* time quorum expires */
418 syncsite = ubeacon_AmSyncSite();
419 startTime = FT_ApproxTime();
421 * Don't waste time using mult Rx calls if there are no connections out there
425 multi_Rx(connections, i) {
426 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
428 temp = FT_ApproxTime(); /* now, more or less */
429 ts = servers[multi_i];
430 ts->lastBeaconSent = temp;
432 /* note that the vote time (the return code) represents the time
433 * the vote was computed, *not* the time the vote expires. We compute
434 * the latter down below if we got enough votes to go with */
436 ts->lastVoteTime = code;
437 if (code < oldestYesVote)
438 oldestYesVote = code;
443 yesVotes++; /* the extra epsilon */
444 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
445 ts->beaconSinceDown = 1;
446 ubik_dprint("yes vote from host %s\n",
447 afs_inet_ntoa_r(ts->addr[0], hoststr));
448 } else if (code == 0) {
449 ts->lastVoteTime = temp;
451 ts->beaconSinceDown = 1;
452 ubik_dprint("no vote from %s\n",
453 afs_inet_ntoa_r(ts->addr[0], hoststr));
454 } else if (code < 0) {
456 ts->beaconSinceDown = 0;
457 urecovery_LostServer();
458 ubik_dprint("time out from %s\n",
459 afs_inet_ntoa_r(ts->addr[0], hoststr));
464 /* now call our own voter module to see if we'll vote for ourself. Note that
465 * the same restrictions apply for our voting for ourself as for our voting
466 * for anyone else. */
467 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
468 &ubik_dbase->version, &ttid);
472 yesVotes++; /* extra epsilon */
473 if (i < oldestYesVote)
476 #if defined(UBIK_PAUSE)
477 ubik_dbase->flags &= ~DBVOTING;
478 #endif /* UBIK_PAUSE */
480 /* now decide if we have enough votes to become sync site.
481 * Note that we can still get enough votes even if we didn't for ourself. */
482 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
483 if (!ubik_amSyncSite)
484 ubik_dprint("Ubik: I am the sync site\n");
486 syncSiteUntil = oldestYesVote + SMALLTIME;
487 #ifndef AFS_PTHREAD_ENV
488 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
489 this may be a spurious signal call -- sjenkins */
490 LWP_NoYieldSignal(&ubik_amSyncSite);
494 ubik_dprint("Ubik: I am no longer the sync site\n");
496 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
504 * \brief Verify that a given IP addresses does actually exist on this machine.
506 * \param ame the pointer to my IP address specified in the
508 * \param aservers an array containing IP
509 * addresses of remote ubik servers. The array is
510 * terminated by a zero address.
512 * Algorithm : Verify that my IP addresses \p ame does actually exist
513 * on this machine. If any of my IP addresses are there
514 * in the remote server list \p aserver, remove them from
515 * this list. Update global variable \p ubik_host[] with
518 * \return 0 on success, non-zero on failure
521 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
522 afs_uint32 aservers[]) {
523 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
524 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
526 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
530 totalServers = info->numServers;
531 else { /* count the number of servers */
532 for (totalServers = 0, servList = aservers; *servList; servList++)
536 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
538 * Find addresses we are supposed to register as per the netrestrict file
539 * if it exists, else just register all the addresses we find on this
540 * host as returned by rx_getAllAddr (in NBO)
544 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
545 AFSDIR_SERVER_NETINFO_FILEPATH,
546 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
548 ubik_print("ubik: Can't register any valid addresses:%s\n",
550 ubik_print("Aborting..\n");
555 /* get all my interface addresses in net byte order */
556 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
559 if (count <= 0) { /* no address found */
560 ubik_print("ubik: No network addresses found, aborting..");
564 /* verify that the My-address passed in by ubik is correct */
565 for (j = 0, found = 0; j < count; j++) {
566 if (*ame == myAddr[j]) { /* both in net byte order */
573 ubik_print("ubik: primary address %s does not exist\n",
574 afs_inet_ntoa_r(*ame, hoststr));
575 /* if we had the result of rx_getAllAddr already, avoid subverting
576 * the "is gethostbyname(gethostname()) us" check. If we're
577 * using NetInfo/NetRestrict, we assume they have enough clue
578 * to avoid that big hole in their foot from the loaded gun. */
580 /* take the address we did get, then see if ame was masked */
582 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
583 if (tcount <= 0) { /* no address found */
584 ubik_print("ubik: No network addresses found, aborting..");
588 /* verify that the My-address passed in by ubik is correct */
589 for (j = 0, found = 0; j < tcount; j++) {
590 if (*ame == myAddr2[j]) { /* both in net byte order */
600 /* if any of my addresses are there in serverList, then
601 ** use that as my primary addresses : the higher level
602 ** application screwed up in dealing with multihomed concepts
604 for (j = 0, found = 0; j < count; j++) {
605 for (i = 0; i < totalServers; i++) {
607 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
609 tmpAddr = aservers[i];
610 if (myAddr[j] == tmpAddr) {
619 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
622 /* get rid of servers which were purged because all
623 ** those interface addresses are myself
625 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
626 /* find the first zero entry from the beginning */
627 for (; (start < end) && (aservers[start]); start++);
629 /* find the last non-zero entry from the end */
630 for (; (end >= 0) && (!aservers[end]); end--);
632 /* if there is nothing more to purge, exit from loop */
637 aservers[start] = aservers[end];
638 aservers[end] = 0; /* this entry was moved */
642 /* update all my addresses in ubik_host in such a way
643 ** that ubik_host[0] has the primary address
646 for (j = 0, i = 1; j < count; j++)
647 if (*ame != myAddr[j])
648 ubik_host[i++] = myAddr[j];
650 return 0; /* return success */
655 * \brief Exchange IP address information with remote servers.
657 * \param ubik_host an array containing all my IP addresses.
659 * Algorithm : Do an RPC to all remote ubik servers infroming them
660 * about my IP addresses. Get their IP addresses and
661 * update my linked list of ubik servers \p ubik_servers
663 * \return 0 on success, non-zero on failure
666 updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
668 int j, count, code = 0;
669 UbikInterfaceAddr inAddr, outAddr;
670 struct rx_connection *conns[MAXSERVERS];
671 struct ubik_server *ts, *server[MAXSERVERS];
675 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
676 conns[count] = ts->disk_rxcid;
681 /* inform all other servers only if there are more than one
682 * database servers in the cell */
686 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
687 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
690 /* do the multi-RX RPC to all other servers */
691 multi_Rx(conns, count) {
692 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
693 ts = server[multi_i]; /* reply received from this server */
695 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
697 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
698 ubik_print("ubik:Two primary addresses for same server \
700 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
702 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
703 ts->addr[j] = htonl(outAddr.hostAddr[j]);
705 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
707 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
708 afs_inet_ntoa_r(ts->addr[0], hoststr));
709 } else if (multi_error == UBADHOST) {
710 code = UBADHOST; /* remote CellServDB inconsistency */
711 ubik_print("Inconsistent Cell Info on server: ");
712 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
713 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
716 ts->up = 0; /* mark the remote server as down */