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>
14 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <netinet/in.h>
33 #include <rx/rx_multi.h>
34 #include <afs/cellconfig.h>
36 #include <afs/afsutil.h>
37 #include <afs/netutils.h>
40 #define UBIK_INTERNALS
44 /*! \name statics used to determine if we're the sync site */
45 static afs_int32 syncSiteUntil = 0; /*!< valid only if amSyncSite */
46 int ubik_amSyncSite = 0; /*!< flag telling if I'm sync site */
47 static int nServers; /*!< total number of servers */
48 static char amIMagic = 0; /*!< is this host the magic host */
49 char amIClone = 0; /*!< is this a clone which doesn't vote */
50 static char ubik_singleServer = 0;
52 int (*ubik_CRXSecurityProc) (void *rock, struct rx_securityClass **,
54 void *ubik_CRXSecurityRock;
55 afs_int32 ubikSecIndex;
56 struct rx_securityClass *ubikSecClass;
57 static int ubeacon_InitServerListCommon(afs_uint32 ame,
58 struct afsconf_cell *info,
60 afs_uint32 aservers[]);
61 static int verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
62 afs_uint32 aservers[]);
63 static int updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR]);
67 * Module responsible for both deciding if we're currently the sync site,
68 * and keeping collecting votes so as to stay sync site.
70 * The basic module contacts all of the servers it can, trying to get them to vote
71 * for this server for sync site. The vote request message (called a beacon message)
72 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
73 * receiving sites to know how long the sync site guarantee is made for.
75 * Each of these beacon messages is thus both a declaration of how long this site will
76 * remain sync site, and an attempt to extend that time by collecting votes for a later
77 * sync site extension.
79 * The voting module is responsible for choosing a reasonable time until which it promises
80 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
81 * the interface (perhaps it should be?) but is instead a compile time constant that both
84 * The beacon and vote modules work intimately together; the vote module decides how long
85 * it should promise the beacon module its vote, and the beacon module takes all of these
86 * votes and decides for how long it is the synchronization site.
89 /*! \brief procedure called from debug rpc call to get this module's state for debugging */
91 ubeacon_Debug(struct ubik_debug *aparm)
93 /* fill in beacon's state fields in the ubik_debug structure */
94 aparm->syncSiteUntil = syncSiteUntil;
95 aparm->nServers = nServers;
99 * \brief Procedure that determines whether this site has enough current votes to remain sync site.
101 * Called from higher-level modules (everything but the vote module).
103 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact()
104 * light-weight process, haven't expired. We're sync site as long as a majority of the
105 * servers in existence have promised us unexpired guarantees. The variable #ubik_syncSiteUntil
106 * contains the time at which the latest of the majority of the sync site guarantees expires
107 * (if the variable #ubik_amSyncSite is true)
108 * This module also calls up to the recovery module if it thinks that the recovery module
109 * may have to pick up a new database (which offucr sif [sic] we lose the sync site votes).
111 * \return 1 if local site is the sync site
112 * \return 0 if sync site is elsewhere
115 ubeacon_AmSyncSite(void)
120 /* special case for fast startup */
121 if (nServers == 1 && !amIClone) {
122 return 1; /* one guy is always the sync site */
125 if (ubik_amSyncSite == 0 || amIClone)
126 rcode = 0; /* if I don't think I'm the sync site, say so */
128 now = FT_ApproxTime();
129 if (syncSiteUntil <= now) { /* if my votes have expired, say so */
131 ubik_dprint("Ubik: I am no longer the sync site\n");
135 rcode = 1; /* otherwise still have the required votes */
139 urecovery_ResetState(); /* force recovery to re-execute */
140 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
145 * \see ubeacon_InitServerListCommon()
148 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
153 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
158 * \param ame "address of me"
159 * \param aservers list of other servers
161 * \see ubeacon_InitServerListCommon()
164 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
169 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
175 * \brief setup server list
177 * \param ame "address of me"
178 * \param aservers list of other servers
180 * called only at initialization to set up the list of servers to
181 * contact for votes. Just creates the server structure.
183 * The "magic" host is the one with the lowest internet address. It is
184 * magic because its vote counts epsilon more than the others. This acts
185 * as a tie-breaker when we have an even number of hosts in the system.
186 * For example, if the "magic" host is up in a 2 site system, then it
187 * is sync site. Without the magic host hack, if anyone crashed in a 2
188 * site system, we'd be out of business.
190 * \note There are two connections in every server structure, one for
191 * vote calls (which must always go through quickly) and one for database
192 * operations, which are subject to waiting for locks. If we used only
193 * one, the votes would sometimes get held up behind database operations,
194 * and the sync site guarantees would timeout even though the host would be
195 * up for communication.
197 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
200 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
201 char clones[], afs_uint32 aservers[])
203 struct ubik_server *ts;
208 struct ubik_server *magicServer;
210 /* verify that the addresses passed in are correct */
211 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
214 /* get the security index to use, if we can */
215 if (ubik_CRXSecurityProc) {
216 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &ubikSecClass,
221 /* don't have sec module yet */
223 ubikSecClass = rxnull_NewClientSecurityObject();
225 magicHost = ntohl(ame); /* do comparisons in host order */
226 magicServer = (struct ubik_server *)0;
229 for (i = 0; i < info->numServers; i++) {
230 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
231 ntohl((afs_uint32) ame)) {
240 for (i = 0; i < info->numServers; i++) {
243 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
244 memset(ts, 0, sizeof(struct ubik_server));
245 ts->next = ubik_servers;
247 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
252 || ntohl((afs_uint32) ts->addr[0]) <
253 (afs_uint32) magicHost) {
254 magicHost = ntohl(ts->addr[0]);
261 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
262 ubik_callPortal, VOTE_SERVICE_ID,
263 ubikSecClass, ubikSecIndex);
266 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
267 ubik_callPortal, DISK_SERVICE_ID,
268 ubikSecClass, ubikSecIndex);
273 while ((servAddr = *aservers++)) {
275 return UNHOSTS; /* too many hosts */
276 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
277 memset(ts, 0, sizeof(struct ubik_server));
278 ts->next = ubik_servers;
280 ts->addr[0] = servAddr; /* primary address in net byte order */
281 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for vote reqs */
282 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for disk reqs */
283 ts->isClone = 0; /* don't know about clones */
285 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
286 magicHost = ntohl(servAddr);
293 magicServer->magic = 1; /* remember for when counting votes */
295 if (!amIClone && !magicServer)
299 ++nServers; /* count this server as well as the remotes */
301 nServers = i + 1; /* count this server as well as the remotes */
303 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
304 /* send addrs to all other servers */
305 code = updateUbikNetworkAddress(ubik_host);
309 /* Shoud we set some defaults for RX??
311 r_nRetries = (RPCTIMEOUT/r_retryInterval);
314 if (!ubik_servers) /* special case 1 server */
315 ubik_singleServer = 1;
316 if (nServers == 1 && !amIClone) {
317 ubik_amSyncSite = 1; /* let's start as sync site */
318 syncSiteUntil = 0x7fffffff; /* and be it quite a while */
321 if (nServers == 1) /* special case 1 server */
322 ubik_singleServer = 1;
325 if (ubik_singleServer) {
326 if (!ubik_amSyncSite)
327 ubik_dprint("Ubik: I am the sync site - 1 server\n");
329 syncSiteUntil = 0x7fffffff; /* quite a while */
335 * \brief main lwp loop for code that sends out beacons.
337 * This code only runs while we're sync site or we want to be the sync site.
338 * It runs in its very own light-weight process.
341 ubeacon_Interact(void *dummy)
345 struct rx_connection *connections[MAXSERVERS];
346 struct ubik_server *servers[MAXSERVERS];
348 struct ubik_server *ts;
349 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
350 struct ubik_tid ttid;
353 /* loop forever getting votes */
354 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
357 /* don't wakeup more than every POLLTIME seconds */
358 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
359 /* don't sleep if last collection phase took too long (probably timed someone out ) */
365 #ifdef AFS_PTHREAD_ENV
366 code = select(0, 0, 0, 0, &tt);
368 code = IOMGR_Select(0, 0, 0, 0, &tt);
373 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
375 if (ubik_singleServer)
376 continue; /* special-case 1 server for speedy startup */
378 if (!uvote_ShouldIRun())
379 continue; /* if voter has heard from a better candidate than us, don't bother running */
381 /* otherwise we should run for election, or we're the sync site (and have already won);
382 * send out the beacon packets */
383 /* build list of all up hosts (noticing dead hosts are running again
384 * is a task for the recovery module, not the beacon module), and
385 * prepare to send them an r multi-call containing the beacon message */
386 i = 0; /* collect connections */
387 for (ts = ubik_servers; ts; ts = ts->next) {
388 if (ts->up && ts->addr[0] != ubik_host[0]) {
390 connections[i++] = ts->vote_rxcid;
393 servers[i] = (struct ubik_server *)0; /* end of list */
394 /* note that we assume in the vote module that we'll always get at least BIGTIME
395 * seconds of vote from anyone who votes for us, which means we can conservatively
396 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
397 /* this next is essentially an expansion of rgen's ServBeacon routine */
399 ttid.epoch = ubik_epochTime;
400 if (ubik_dbase->flags & DBWRITING) {
402 * if a write is in progress, we have to send the writeTidCounter
403 * which holds the tid counter of the write transaction , and not
404 * send the tidCounter value which holds the tid counter of the
407 ttid.counter = ubik_dbase->writeTidCounter;
409 ttid.counter = ubik_dbase->tidCounter + 1;
410 #if defined(UBIK_PAUSE)
411 ubik_dbase->flags |= DBVOTING;
412 #endif /* UBIK_PAUSE */
414 /* now analyze return codes, counting up our votes */
415 yesVotes = 0; /* count how many to ensure we have quorum */
416 oldestYesVote = 0x7fffffff; /* time quorum expires */
417 syncsite = ubeacon_AmSyncSite();
418 startTime = FT_ApproxTime();
420 * Don't waste time using mult Rx calls if there are no connections out there
424 multi_Rx(connections, i) {
425 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
427 temp = FT_ApproxTime(); /* now, more or less */
428 ts = servers[multi_i];
429 ts->lastBeaconSent = temp;
431 /* note that the vote time (the return code) represents the time
432 * the vote was computed, *not* the time the vote expires. We compute
433 * the latter down below if we got enough votes to go with */
435 ts->lastVoteTime = code;
436 if (code < oldestYesVote)
437 oldestYesVote = code;
442 yesVotes++; /* the extra epsilon */
443 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
444 ts->beaconSinceDown = 1;
445 ubik_dprint("yes vote from host %s\n",
446 afs_inet_ntoa_r(ts->addr[0], hoststr));
447 } else if (code == 0) {
448 ts->lastVoteTime = temp;
450 ts->beaconSinceDown = 1;
451 ubik_dprint("no vote from %s\n",
452 afs_inet_ntoa_r(ts->addr[0], hoststr));
453 } else if (code < 0) {
455 ts->beaconSinceDown = 0;
456 urecovery_LostServer();
457 ubik_dprint("time out from %s\n",
458 afs_inet_ntoa_r(ts->addr[0], hoststr));
463 /* now call our own voter module to see if we'll vote for ourself. Note that
464 * the same restrictions apply for our voting for ourself as for our voting
465 * for anyone else. */
466 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
467 &ubik_dbase->version, &ttid);
471 yesVotes++; /* extra epsilon */
472 if (i < oldestYesVote)
475 #if defined(UBIK_PAUSE)
476 ubik_dbase->flags &= ~DBVOTING;
477 #endif /* UBIK_PAUSE */
479 /* now decide if we have enough votes to become sync site.
480 * Note that we can still get enough votes even if we didn't for ourself. */
481 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
482 if (!ubik_amSyncSite)
483 ubik_dprint("Ubik: I am the sync site\n");
485 syncSiteUntil = oldestYesVote + SMALLTIME;
486 #ifndef AFS_PTHREAD_ENV
487 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
488 this may be a spurious signal call -- sjenkins */
489 LWP_NoYieldSignal(&ubik_amSyncSite);
493 ubik_dprint("Ubik: I am no longer the sync site\n");
495 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
503 * \brief Verify that a given IP addresses does actually exist on this machine.
505 * \param ame the pointer to my IP address specified in the
507 * \param aservers an array containing IP
508 * addresses of remote ubik servers. The array is
509 * terminated by a zero address.
511 * Algorithm : Verify that my IP addresses \p ame does actually exist
512 * on this machine. If any of my IP addresses are there
513 * in the remote server list \p aserver, remove them from
514 * this list. Update global variable \p ubik_host[] with
517 * \return 0 on success, non-zero on failure
520 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
521 afs_uint32 aservers[]) {
522 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
523 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
525 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
529 totalServers = info->numServers;
530 else { /* count the number of servers */
531 for (totalServers = 0, servList = aservers; *servList; servList++)
535 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
537 * Find addresses we are supposed to register as per the netrestrict file
538 * if it exists, else just register all the addresses we find on this
539 * host as returned by rx_getAllAddr (in NBO)
543 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
544 AFSDIR_SERVER_NETINFO_FILEPATH,
545 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
547 ubik_print("ubik: Can't register any valid addresses:%s\n",
549 ubik_print("Aborting..\n");
554 /* get all my interface addresses in net byte order */
555 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
558 if (count <= 0) { /* no address found */
559 ubik_print("ubik: No network addresses found, aborting..");
563 /* verify that the My-address passed in by ubik is correct */
564 for (j = 0, found = 0; j < count; j++) {
565 if (*ame == myAddr[j]) { /* both in net byte order */
572 ubik_print("ubik: primary address %s does not exist\n",
573 afs_inet_ntoa_r(*ame, hoststr));
574 /* if we had the result of rx_getAllAddr already, avoid subverting
575 * the "is gethostbyname(gethostname()) us" check. If we're
576 * using NetInfo/NetRestrict, we assume they have enough clue
577 * to avoid that big hole in their foot from the loaded gun. */
579 /* take the address we did get, then see if ame was masked */
581 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
582 if (tcount <= 0) { /* no address found */
583 ubik_print("ubik: No network addresses found, aborting..");
587 /* verify that the My-address passed in by ubik is correct */
588 for (j = 0, found = 0; j < tcount; j++) {
589 if (*ame == myAddr2[j]) { /* both in net byte order */
599 /* if any of my addresses are there in serverList, then
600 ** use that as my primary addresses : the higher level
601 ** application screwed up in dealing with multihomed concepts
603 for (j = 0, found = 0; j < count; j++) {
604 for (i = 0; i < totalServers; i++) {
606 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
608 tmpAddr = aservers[i];
609 if (myAddr[j] == tmpAddr) {
618 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
621 /* get rid of servers which were purged because all
622 ** those interface addresses are myself
624 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
625 /* find the first zero entry from the beginning */
626 for (; (start < end) && (aservers[start]); start++);
628 /* find the last non-zero entry from the end */
629 for (; (end >= 0) && (!aservers[end]); end--);
631 /* if there is nothing more to purge, exit from loop */
636 aservers[start] = aservers[end];
637 aservers[end] = 0; /* this entry was moved */
641 /* update all my addresses in ubik_host in such a way
642 ** that ubik_host[0] has the primary address
645 for (j = 0, i = 1; j < count; j++)
646 if (*ame != myAddr[j])
647 ubik_host[i++] = myAddr[j];
649 return 0; /* return success */
654 * \brief Exchange IP address information with remote servers.
656 * \param ubik_host an array containing all my IP addresses.
658 * Algorithm : Do an RPC to all remote ubik servers infroming them
659 * about my IP addresses. Get their IP addresses and
660 * update my linked list of ubik servers \p ubik_servers
662 * \return 0 on success, non-zero on failure
665 updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
667 int j, count, code = 0;
668 UbikInterfaceAddr inAddr, outAddr;
669 struct rx_connection *conns[MAXSERVERS];
670 struct ubik_server *ts, *server[MAXSERVERS];
674 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
675 conns[count] = ts->disk_rxcid;
680 /* inform all other servers only if there are more than one
681 * database servers in the cell */
685 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
686 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
689 /* do the multi-RX RPC to all other servers */
690 multi_Rx(conns, count) {
691 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
692 ts = server[multi_i]; /* reply received from this server */
694 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
696 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
697 ubik_print("ubik:Two primary addresses for same server \
699 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
701 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
702 ts->addr[j] = htonl(outAddr.hostAddr[j]);
704 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
706 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
707 afs_inet_ntoa_r(ts->addr[0], hoststr));
708 } else if (multi_error == UBADHOST) {
709 code = UBADHOST; /* remote CellServDB inconsistency */
710 ubik_print("Inconsistent Cell Info on server: ");
711 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
712 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
715 ts->up = 0; /* mark the remote server as down */