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>
35 #include <rx/rx_multi.h>
36 #include <afs/cellconfig.h>
38 #include <afs/afsutil.h>
39 #include <afs/netutils.h>
42 #define UBIK_INTERNALS
46 /*! \name statics used to determine if we're the sync site */
47 static afs_int32 syncSiteUntil = 0; /*!< valid only if amSyncSite */
48 int ubik_amSyncSite = 0; /*!< flag telling if I'm sync site */
49 static int nServers; /*!< total number of servers */
50 static char amIMagic = 0; /*!< is this host the magic host */
51 char amIClone = 0; /*!< is this a clone which doesn't vote */
52 static char ubik_singleServer = 0;
54 int (*ubik_CRXSecurityProc) (void *rock, struct rx_securityClass **,
56 void *ubik_CRXSecurityRock;
57 afs_int32 ubikSecIndex;
58 struct rx_securityClass *ubikSecClass;
59 static int ubeacon_InitServerListCommon(afs_uint32 ame,
60 struct afsconf_cell *info,
62 afs_uint32 aservers[]);
63 static int verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
64 afs_uint32 aservers[]);
65 static int updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR]);
69 * Module responsible for both deciding if we're currently the sync site,
70 * and keeping collecting votes so as to stay sync site.
72 * The basic module contacts all of the servers it can, trying to get them to vote
73 * for this server for sync site. The vote request message (called a beacon message)
74 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
75 * receiving sites to know how long the sync site guarantee is made for.
77 * Each of these beacon messages is thus both a declaration of how long this site will
78 * remain sync site, and an attempt to extend that time by collecting votes for a later
79 * sync site extension.
81 * The voting module is responsible for choosing a reasonable time until which it promises
82 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
83 * the interface (perhaps it should be?) but is instead a compile time constant that both
86 * The beacon and vote modules work intimately together; the vote module decides how long
87 * it should promise the beacon module its vote, and the beacon module takes all of these
88 * votes and decides for how long it is the synchronization site.
91 /*! \brief procedure called from debug rpc call to get this module's state for debugging */
93 ubeacon_Debug(struct ubik_debug *aparm)
95 /* fill in beacon's state fields in the ubik_debug structure */
96 aparm->syncSiteUntil = syncSiteUntil;
97 aparm->nServers = nServers;
101 * \brief Procedure that determines whether this site has enough current votes to remain sync site.
103 * Called from higher-level modules (everything but the vote module).
105 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact()
106 * light-weight process, haven't expired. We're sync site as long as a majority of the
107 * servers in existence have promised us unexpired guarantees. The variable #ubik_syncSiteUntil
108 * contains the time at which the latest of the majority of the sync site guarantees expires
109 * (if the variable #ubik_amSyncSite is true)
110 * This module also calls up to the recovery module if it thinks that the recovery module
111 * may have to pick up a new database (which offucr sif [sic] we lose the sync site votes).
113 * \return 1 if local site is the sync site
114 * \return 0 if sync site is elsewhere
117 ubeacon_AmSyncSite(void)
122 /* special case for fast startup */
123 if (nServers == 1 && !amIClone) {
124 return 1; /* one guy is always the sync site */
127 if (ubik_amSyncSite == 0 || amIClone)
128 rcode = 0; /* if I don't think I'm the sync site, say so */
130 now = FT_ApproxTime();
131 if (syncSiteUntil <= now) { /* if my votes have expired, say so */
133 ubik_dprint("Ubik: I am no longer the sync site\n");
137 rcode = 1; /* otherwise still have the required votes */
141 urecovery_ResetState(); /* force recovery to re-execute */
142 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
147 * \see ubeacon_InitServerListCommon()
150 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
155 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
160 * \param ame "address of me"
161 * \param aservers list of other servers
163 * \see ubeacon_InitServerListCommon()
166 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
171 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
177 ubeacon_InitSecurityClass(void)
180 /* get the security index to use, if we can */
181 if (ubik_CRXSecurityProc) {
182 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &ubikSecClass,
187 /* don't have sec module yet */
189 ubikSecClass = rxnull_NewClientSecurityObject();
194 ubeacon_ReinitServer(struct ubik_server *ts)
196 if (!afsconf_UpToDate(ubik_CRXSecurityRock)) {
197 struct rx_connection *disk_rxcid;
198 struct rx_connection *vote_rxcid;
199 struct rx_connection *tmp;
200 ubeacon_InitSecurityClass();
202 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->disk_rxcid)),
203 ubik_callPortal, DISK_SERVICE_ID,
204 ubikSecClass, ubikSecIndex);
206 tmp = ts->disk_rxcid;
207 ts->disk_rxcid = disk_rxcid;
208 rx_PutConnection(tmp);
211 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->vote_rxcid)),
212 ubik_callPortal, VOTE_SERVICE_ID,
213 ubikSecClass, ubikSecIndex);
215 tmp = ts->vote_rxcid;
216 ts->vote_rxcid = vote_rxcid;
217 rx_PutConnection(tmp);
223 * \brief setup server list
225 * \param ame "address of me"
226 * \param aservers list of other servers
228 * called only at initialization to set up the list of servers to
229 * contact for votes. Just creates the server structure.
231 * The "magic" host is the one with the lowest internet address. It is
232 * magic because its vote counts epsilon more than the others. This acts
233 * as a tie-breaker when we have an even number of hosts in the system.
234 * For example, if the "magic" host is up in a 2 site system, then it
235 * is sync site. Without the magic host hack, if anyone crashed in a 2
236 * site system, we'd be out of business.
238 * \note There are two connections in every server structure, one for
239 * vote calls (which must always go through quickly) and one for database
240 * operations, which are subject to waiting for locks. If we used only
241 * one, the votes would sometimes get held up behind database operations,
242 * and the sync site guarantees would timeout even though the host would be
243 * up for communication.
245 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
248 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
249 char clones[], afs_uint32 aservers[])
251 struct ubik_server *ts;
256 struct ubik_server *magicServer;
258 /* verify that the addresses passed in are correct */
259 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
262 ubeacon_InitSecurityClass();
264 magicHost = ntohl(ame); /* do comparisons in host order */
265 magicServer = (struct ubik_server *)0;
268 for (i = 0; i < info->numServers; i++) {
269 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
270 ntohl((afs_uint32) ame)) {
279 for (i = 0; i < info->numServers; i++) {
282 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
283 memset(ts, 0, sizeof(struct ubik_server));
284 ts->next = ubik_servers;
286 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
291 || ntohl((afs_uint32) ts->addr[0]) <
292 (afs_uint32) magicHost) {
293 magicHost = ntohl(ts->addr[0]);
300 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
301 ubik_callPortal, VOTE_SERVICE_ID,
302 ubikSecClass, ubikSecIndex);
305 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
306 ubik_callPortal, DISK_SERVICE_ID,
307 ubikSecClass, ubikSecIndex);
312 while ((servAddr = *aservers++)) {
314 return UNHOSTS; /* too many hosts */
315 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
316 memset(ts, 0, sizeof(struct ubik_server));
317 ts->next = ubik_servers;
319 ts->addr[0] = servAddr; /* primary address in net byte order */
320 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for vote reqs */
321 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for disk reqs */
322 ts->isClone = 0; /* don't know about clones */
324 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
325 magicHost = ntohl(servAddr);
332 magicServer->magic = 1; /* remember for when counting votes */
334 if (!amIClone && !magicServer)
338 ++nServers; /* count this server as well as the remotes */
340 nServers = i + 1; /* count this server as well as the remotes */
342 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
343 /* send addrs to all other servers */
344 code = updateUbikNetworkAddress(ubik_host);
348 /* Shoud we set some defaults for RX??
350 r_nRetries = (RPCTIMEOUT/r_retryInterval);
353 if (!ubik_servers) /* special case 1 server */
354 ubik_singleServer = 1;
355 if (nServers == 1 && !amIClone) {
356 ubik_amSyncSite = 1; /* let's start as sync site */
357 syncSiteUntil = 0x7fffffff; /* and be it quite a while */
360 if (nServers == 1) /* special case 1 server */
361 ubik_singleServer = 1;
364 if (ubik_singleServer) {
365 if (!ubik_amSyncSite)
366 ubik_dprint("Ubik: I am the sync site - 1 server\n");
368 syncSiteUntil = 0x7fffffff; /* quite a while */
374 * \brief main lwp loop for code that sends out beacons.
376 * This code only runs while we're sync site or we want to be the sync site.
377 * It runs in its very own light-weight process.
380 ubeacon_Interact(void *dummy)
384 struct rx_connection *connections[MAXSERVERS];
385 struct ubik_server *servers[MAXSERVERS];
387 struct ubik_server *ts;
388 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
389 struct ubik_tid ttid;
392 /* loop forever getting votes */
393 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
396 /* don't wakeup more than every POLLTIME seconds */
397 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
398 /* don't sleep if last collection phase took too long (probably timed someone out ) */
404 #ifdef AFS_PTHREAD_ENV
405 code = select(0, 0, 0, 0, &tt);
407 code = IOMGR_Select(0, 0, 0, 0, &tt);
412 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
414 if (ubik_singleServer)
415 continue; /* special-case 1 server for speedy startup */
417 if (!uvote_ShouldIRun())
418 continue; /* if voter has heard from a better candidate than us, don't bother running */
420 /* otherwise we should run for election, or we're the sync site (and have already won);
421 * send out the beacon packets */
422 /* build list of all up hosts (noticing dead hosts are running again
423 * is a task for the recovery module, not the beacon module), and
424 * prepare to send them an r multi-call containing the beacon message */
425 i = 0; /* collect connections */
426 for (ts = ubik_servers; ts; ts = ts->next) {
427 if (ts->up && ts->addr[0] != ubik_host[0]) {
429 connections[i++] = ts->vote_rxcid;
432 servers[i] = (struct ubik_server *)0; /* end of list */
433 /* note that we assume in the vote module that we'll always get at least BIGTIME
434 * seconds of vote from anyone who votes for us, which means we can conservatively
435 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
436 /* this next is essentially an expansion of rgen's ServBeacon routine */
438 ttid.epoch = ubik_epochTime;
439 if (ubik_dbase->flags & DBWRITING) {
441 * if a write is in progress, we have to send the writeTidCounter
442 * which holds the tid counter of the write transaction , and not
443 * send the tidCounter value which holds the tid counter of the
446 ttid.counter = ubik_dbase->writeTidCounter;
448 ttid.counter = ubik_dbase->tidCounter + 1;
450 /* now analyze return codes, counting up our votes */
451 yesVotes = 0; /* count how many to ensure we have quorum */
452 oldestYesVote = 0x7fffffff; /* time quorum expires */
453 syncsite = ubeacon_AmSyncSite();
454 startTime = FT_ApproxTime();
456 * Don't waste time using mult Rx calls if there are no connections out there
460 multi_Rx(connections, i) {
461 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
463 temp = FT_ApproxTime(); /* now, more or less */
464 ts = servers[multi_i];
465 ts->lastBeaconSent = temp;
467 /* note that the vote time (the return code) represents the time
468 * the vote was computed, *not* the time the vote expires. We compute
469 * the latter down below if we got enough votes to go with */
471 if ((code & ~0xff) == ERROR_TABLE_BASE_RXK) {
472 ubik_dprint("token error %d from host %s\n",
473 code, afs_inet_ntoa_r(ts->addr[0], hoststr));
475 ts->beaconSinceDown = 0;
476 urecovery_LostServer(ts);
478 ts->lastVoteTime = code;
479 if (code < oldestYesVote)
480 oldestYesVote = code;
485 yesVotes++; /* the extra epsilon */
486 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
487 ts->beaconSinceDown = 1;
488 ubik_dprint("yes vote from host %s\n",
489 afs_inet_ntoa_r(ts->addr[0], hoststr));
491 } else if (code == 0) {
492 ts->lastVoteTime = temp;
494 ts->beaconSinceDown = 1;
495 ubik_dprint("no vote from %s\n",
496 afs_inet_ntoa_r(ts->addr[0], hoststr));
497 } else if (code < 0) {
499 ts->beaconSinceDown = 0;
500 urecovery_LostServer(ts);
501 ubik_dprint("time out from %s\n",
502 afs_inet_ntoa_r(ts->addr[0], hoststr));
507 /* now call our own voter module to see if we'll vote for ourself. Note that
508 * the same restrictions apply for our voting for ourself as for our voting
509 * for anyone else. */
510 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
511 &ubik_dbase->version, &ttid);
515 yesVotes++; /* extra epsilon */
516 if (i < oldestYesVote)
520 /* now decide if we have enough votes to become sync site.
521 * Note that we can still get enough votes even if we didn't for ourself. */
522 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
523 if (!ubik_amSyncSite)
524 ubik_dprint("Ubik: I am the sync site\n");
526 syncSiteUntil = oldestYesVote + SMALLTIME;
527 #ifndef AFS_PTHREAD_ENV
528 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
529 this may be a spurious signal call -- sjenkins */
530 LWP_NoYieldSignal(&ubik_amSyncSite);
534 ubik_dprint("Ubik: I am no longer the sync site\n");
536 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
544 * \brief Verify that a given IP addresses does actually exist on this machine.
546 * \param ame the pointer to my IP address specified in the
548 * \param aservers an array containing IP
549 * addresses of remote ubik servers. The array is
550 * terminated by a zero address.
552 * Algorithm : Verify that my IP addresses \p ame does actually exist
553 * on this machine. If any of my IP addresses are there
554 * in the remote server list \p aserver, remove them from
555 * this list. Update global variable \p ubik_host[] with
558 * \return 0 on success, non-zero on failure
561 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
562 afs_uint32 aservers[]) {
563 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
564 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
566 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
570 totalServers = info->numServers;
571 else { /* count the number of servers */
572 for (totalServers = 0, servList = aservers; *servList; servList++)
576 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
578 * Find addresses we are supposed to register as per the netrestrict file
579 * if it exists, else just register all the addresses we find on this
580 * host as returned by rx_getAllAddr (in NBO)
584 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
585 AFSDIR_SERVER_NETINFO_FILEPATH,
586 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
588 ubik_print("ubik: Can't register any valid addresses:%s\n",
590 ubik_print("Aborting..\n");
595 /* get all my interface addresses in net byte order */
596 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
599 if (count <= 0) { /* no address found */
600 ubik_print("ubik: No network addresses found, aborting..");
604 /* verify that the My-address passed in by ubik is correct */
605 for (j = 0, found = 0; j < count; j++) {
606 if (*ame == myAddr[j]) { /* both in net byte order */
613 ubik_print("ubik: primary address %s does not exist\n",
614 afs_inet_ntoa_r(*ame, hoststr));
615 /* if we had the result of rx_getAllAddr already, avoid subverting
616 * the "is gethostbyname(gethostname()) us" check. If we're
617 * using NetInfo/NetRestrict, we assume they have enough clue
618 * to avoid that big hole in their foot from the loaded gun. */
620 /* take the address we did get, then see if ame was masked */
622 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
623 if (tcount <= 0) { /* no address found */
624 ubik_print("ubik: No network addresses found, aborting..");
628 /* verify that the My-address passed in by ubik is correct */
629 for (j = 0, found = 0; j < tcount; j++) {
630 if (*ame == myAddr2[j]) { /* both in net byte order */
640 /* if any of my addresses are there in serverList, then
641 ** use that as my primary addresses : the higher level
642 ** application screwed up in dealing with multihomed concepts
644 for (j = 0, found = 0; j < count; j++) {
645 for (i = 0; i < totalServers; i++) {
647 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
649 tmpAddr = aservers[i];
650 if (myAddr[j] == tmpAddr) {
659 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
662 /* get rid of servers which were purged because all
663 ** those interface addresses are myself
665 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
666 /* find the first zero entry from the beginning */
667 for (; (start < end) && (aservers[start]); start++);
669 /* find the last non-zero entry from the end */
670 for (; (end >= 0) && (!aservers[end]); end--);
672 /* if there is nothing more to purge, exit from loop */
677 aservers[start] = aservers[end];
678 aservers[end] = 0; /* this entry was moved */
682 /* update all my addresses in ubik_host in such a way
683 ** that ubik_host[0] has the primary address
686 for (j = 0, i = 1; j < count; j++)
687 if (*ame != myAddr[j])
688 ubik_host[i++] = myAddr[j];
690 return 0; /* return success */
695 * \brief Exchange IP address information with remote servers.
697 * \param ubik_host an array containing all my IP addresses.
699 * Algorithm : Do an RPC to all remote ubik servers infroming them
700 * about my IP addresses. Get their IP addresses and
701 * update my linked list of ubik servers \p ubik_servers
703 * \return 0 on success, non-zero on failure
706 updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
708 int j, count, code = 0;
709 UbikInterfaceAddr inAddr, outAddr;
710 struct rx_connection *conns[MAXSERVERS];
711 struct ubik_server *ts, *server[MAXSERVERS];
715 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
716 conns[count] = ts->disk_rxcid;
721 /* inform all other servers only if there are more than one
722 * database servers in the cell */
726 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
727 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
730 /* do the multi-RX RPC to all other servers */
731 multi_Rx(conns, count) {
732 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
733 ts = server[multi_i]; /* reply received from this server */
735 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
737 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
738 ubik_print("ubik:Two primary addresses for same server \
740 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
742 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
743 ts->addr[j] = htonl(outAddr.hostAddr[j]);
745 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
747 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
748 afs_inet_ntoa_r(ts->addr[0], hoststr));
749 } else if (multi_error == UBADHOST) {
750 code = UBADHOST; /* remote CellServDB inconsistency */
751 ubik_print("Inconsistent Cell Info on server: ");
752 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
753 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
756 ts->up = 0; /* mark the remote server as down */