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>
19 #include <rx/rx_multi.h>
20 #include <afs/cellconfig.h>
22 #include <afs/afsutil.h>
25 #define UBIK_INTERNALS
29 /* These global variables were used to set the function to use to initialise
30 * the client security layer. They are retained for backwards compatiblity with
31 * legacy callers - the ubik_SetClientSecurityProcs() interface should be used
34 int (*ubik_CRXSecurityProc) (void *rock, struct rx_securityClass **,
36 void *ubik_CRXSecurityRock;
38 /*! \name statics used to determine if we're the sync site */
39 static int nServers; /*!< total number of servers */
40 static char amIMagic = 0; /*!< is this host the magic host */
41 char amIClone = 0; /*!< is this a clone which doesn't vote */
42 static char ubik_singleServer = 0;
44 static int (*secLayerProc) (void *rock, struct rx_securityClass **,
46 static int (*tokenCheckProc) (void *rock) = NULL;
47 static void * securityRock = NULL;
49 afs_int32 ubikSecIndex;
50 struct rx_securityClass *ubikSecClass;
52 /* Values protected by the address lock */
53 struct addr_data addr_globals;
55 /* Values protected by the beacon lock */
56 struct beacon_data beacon_globals;
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[]);
66 * Module responsible for both deciding if we're currently the sync site,
67 * and keeping collecting votes so as to stay sync site.
69 * The basic module contacts all of the servers it can, trying to get them to vote
70 * for this server for sync site. The vote request message (called a beacon message)
71 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
72 * receiving sites to know how long the sync site guarantee is made for.
74 * Each of these beacon messages is thus both a declaration of how long this site will
75 * remain sync site, and an attempt to extend that time by collecting votes for a later
76 * sync site extension.
78 * The voting module is responsible for choosing a reasonable time until which it promises
79 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
80 * the interface (perhaps it should be?) but is instead a compile time constant that both
83 * The beacon and vote modules work intimately together; the vote module decides how long
84 * it should promise the beacon module its vote, and the beacon module takes all of these
85 * votes and decides for how long it is the synchronization site.
88 /*! \brief procedure called from debug rpc call to get this module's state for debugging */
90 ubeacon_Debug(struct ubik_debug *aparm)
92 /* fill in beacon's state fields in the ubik_debug structure */
93 aparm->syncSiteUntil = beacon_globals.syncSiteUntil;
94 aparm->nServers = nServers;
102 /* special case for fast startup */
103 if (nServers == 1 && !amIClone)
104 return 1; /* one guy is always the sync site */
107 if (beacon_globals.ubik_amSyncSite == 0 || amIClone)
108 rcode = 0; /* if I don't think I'm the sync site, say so */
110 now = FT_ApproxTime();
111 if (beacon_globals.syncSiteUntil <= now) { /* if my votes have expired, say so */
112 if (beacon_globals.ubik_amSyncSite)
113 ubik_dprint("Ubik: I am no longer the sync site\n");
114 beacon_globals.ubik_amSyncSite = 0;
117 rcode = 1; /* otherwise still have the required votes */
121 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
126 * \brief Procedure that determines whether this site has enough current votes to remain sync site.
128 * Called from higher-level modules (everything but the vote module).
130 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact()
131 * light-weight process, haven't expired. We're sync site as long as a majority of the
132 * servers in existence have promised us unexpired guarantees. The variable #ubik_syncSiteUntil
133 * contains the time at which the latest of the majority of the sync site guarantees expires
134 * (if the variable #ubik_amSyncSite is true)
135 * This module also calls up to the recovery module if it thinks that the recovery module
136 * may have to pick up a new database (which offucr sif [sic] we lose the sync site votes).
138 * \return 1 if local site is the sync site
139 * \return 0 if sync site is elsewhere
142 ubeacon_AmSyncSite(void)
146 rcode = amSyncSite();
149 urecovery_ResetState();
155 * \see ubeacon_InitServerListCommon()
158 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
163 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
168 * \param ame "address of me"
169 * \param aservers list of other servers
171 * \see ubeacon_InitServerListCommon()
174 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
179 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
184 /* Must be called with address lock held */
186 ubeacon_InitSecurityClass(void)
189 /* get the security index to use, if we can */
191 i = (*secLayerProc) (securityRock, &addr_globals.ubikSecClass, &addr_globals.ubikSecIndex);
192 } else if (ubik_CRXSecurityProc) {
193 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &addr_globals.ubikSecClass,
194 &addr_globals.ubikSecIndex);
198 /* don't have sec module yet */
199 addr_globals.ubikSecIndex = 0;
200 addr_globals.ubikSecClass = rxnull_NewClientSecurityObject();
205 ubeacon_ReinitServer(struct ubik_server *ts)
207 if (tokenCheckProc && !(*tokenCheckProc) (securityRock)) {
208 struct rx_connection *disk_rxcid;
209 struct rx_connection *vote_rxcid;
210 struct rx_connection *tmp;
212 ubeacon_InitSecurityClass();
214 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->disk_rxcid)),
215 ubik_callPortal, DISK_SERVICE_ID,
216 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
218 tmp = ts->disk_rxcid;
219 ts->disk_rxcid = disk_rxcid;
220 rx_PutConnection(tmp);
223 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->vote_rxcid)),
224 ubik_callPortal, VOTE_SERVICE_ID,
225 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
227 tmp = ts->vote_rxcid;
228 ts->vote_rxcid = vote_rxcid;
229 rx_PutConnection(tmp);
236 * \brief setup server list
238 * \param ame "address of me"
239 * \param aservers list of other servers
241 * called only at initialization to set up the list of servers to
242 * contact for votes. Just creates the server structure.
244 * The "magic" host is the one with the lowest internet address. It is
245 * magic because its vote counts epsilon more than the others. This acts
246 * as a tie-breaker when we have an even number of hosts in the system.
247 * For example, if the "magic" host is up in a 2 site system, then it
248 * is sync site. Without the magic host hack, if anyone crashed in a 2
249 * site system, we'd be out of business.
251 * \note There are two connections in every server structure, one for
252 * vote calls (which must always go through quickly) and one for database
253 * operations, which are subject to waiting for locks. If we used only
254 * one, the votes would sometimes get held up behind database operations,
255 * and the sync site guarantees would timeout even though the host would be
256 * up for communication.
258 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
261 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
262 char clones[], afs_uint32 aservers[])
264 struct ubik_server *ts;
269 struct ubik_server *magicServer;
271 /* verify that the addresses passed in are correct */
272 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
275 ubeacon_InitSecurityClass();
277 magicHost = ntohl(ame); /* do comparisons in host order */
278 magicServer = (struct ubik_server *)0;
281 for (i = 0; i < info->numServers; i++) {
282 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
283 ntohl((afs_uint32) ame)) {
292 for (i = 0; i < info->numServers; i++) {
295 ts = calloc(1, sizeof(struct ubik_server));
296 ts->next = ubik_servers;
298 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
303 || ntohl((afs_uint32) ts->addr[0]) <
304 (afs_uint32) magicHost) {
305 magicHost = ntohl(ts->addr[0]);
312 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
313 ubik_callPortal, VOTE_SERVICE_ID,
314 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
317 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
318 ubik_callPortal, DISK_SERVICE_ID,
319 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
324 while ((servAddr = *aservers++)) {
326 return UNHOSTS; /* too many hosts */
327 ts = calloc(1, sizeof(struct ubik_server));
328 ts->next = ubik_servers;
330 ts->addr[0] = servAddr; /* primary address in net byte order */
331 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID,
332 addr_globals.ubikSecClass, addr_globals.ubikSecIndex); /* for vote reqs */
333 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID,
334 addr_globals.ubikSecClass, addr_globals.ubikSecIndex); /* for disk reqs */
335 ts->isClone = 0; /* don't know about clones */
337 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
338 magicHost = ntohl(servAddr);
345 magicServer->magic = 1; /* remember for when counting votes */
347 if (!amIClone && !magicServer)
351 ++nServers; /* count this server as well as the remotes */
353 nServers = i + 1; /* count this server as well as the remotes */
355 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
357 /* Shoud we set some defaults for RX??
359 r_nRetries = (RPCTIMEOUT/r_retryInterval);
362 if (!ubik_servers) /* special case 1 server */
363 ubik_singleServer = 1;
364 if (nServers == 1 && !amIClone) {
365 beacon_globals.ubik_amSyncSite = 1; /* let's start as sync site */
366 beacon_globals.syncSiteUntil = 0x7fffffff; /* and be it quite a while */
369 if (nServers == 1) /* special case 1 server */
370 ubik_singleServer = 1;
373 if (ubik_singleServer) {
374 if (!beacon_globals.ubik_amSyncSite)
375 ubik_dprint("Ubik: I am the sync site - 1 server\n");
376 beacon_globals.ubik_amSyncSite = 1;
377 beacon_globals.syncSiteUntil = 0x7fffffff; /* quite a while */
383 * \brief main lwp loop for code that sends out beacons.
385 * This code only runs while we're sync site or we want to be the sync site.
386 * It runs in its very own light-weight process.
389 ubeacon_Interact(void *dummy)
393 struct rx_connection *connections[MAXSERVERS];
394 struct ubik_server *servers[MAXSERVERS];
396 struct ubik_server *ts;
397 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
398 struct ubik_tid ttid;
399 struct ubik_version tversion;
402 afs_pthread_setname_self("beacon");
404 /* loop forever getting votes */
405 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
408 /* don't wakeup more than every POLLTIME seconds */
409 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
410 /* don't sleep if last collection phase took too long (probably timed someone out ) */
416 #ifdef AFS_PTHREAD_ENV
417 code = select(0, 0, 0, 0, &tt);
419 code = IOMGR_Select(0, 0, 0, 0, &tt);
424 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
426 if (ubik_singleServer)
427 continue; /* special-case 1 server for speedy startup */
429 if (!uvote_ShouldIRun())
430 continue; /* if voter has heard from a better candidate than us, don't bother running */
432 /* otherwise we should run for election, or we're the sync site (and have already won);
433 * send out the beacon packets */
434 /* build list of all up hosts (noticing dead hosts are running again
435 * is a task for the recovery module, not the beacon module), and
436 * prepare to send them an r multi-call containing the beacon message */
437 i = 0; /* collect connections */
440 for (ts = ubik_servers; ts; ts = ts->next) {
441 if (ts->up && ts->addr[0] != ubik_host[0]) {
443 connections[i++] = ts->vote_rxcid;
448 servers[i] = (struct ubik_server *)0; /* end of list */
449 /* note that we assume in the vote module that we'll always get at least BIGTIME
450 * seconds of vote from anyone who votes for us, which means we can conservatively
451 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
452 /* this next is essentially an expansion of rgen's ServBeacon routine */
455 ttid.epoch = version_globals.ubik_epochTime;
456 if (ubik_dbase->flags & DBWRITING) {
458 * if a write is in progress, we have to send the writeTidCounter
459 * which holds the tid counter of the write transaction , and not
460 * send the tidCounter value which holds the tid counter of the
463 ttid.counter = ubik_dbase->writeTidCounter;
465 ttid.counter = ubik_dbase->tidCounter + 1;
466 tversion.epoch = ubik_dbase->version.epoch;
467 tversion.counter = ubik_dbase->version.counter;
470 /* now analyze return codes, counting up our votes */
471 yesVotes = 0; /* count how many to ensure we have quorum */
472 oldestYesVote = 0x7fffffff; /* time quorum expires */
473 syncsite = amSyncSite();
475 /* Ok to use the DB lock here since we aren't sync site */
477 urecovery_ResetState();
480 startTime = FT_ApproxTime();
482 * Don't waste time using mult Rx calls if there are no connections out there
486 multi_Rx(connections, i) {
487 multi_VOTE_Beacon(syncsite, startTime, &tversion,
489 temp = FT_ApproxTime(); /* now, more or less */
490 ts = servers[multi_i];
492 ts->lastBeaconSent = temp;
494 /* note that the vote time (the return code) represents the time
495 * the vote was computed, *not* the time the vote expires. We compute
496 * the latter down below if we got enough votes to go with */
498 if ((code & ~0xff) == ERROR_TABLE_BASE_RXK) {
499 ubik_dprint("token error %d from host %s\n",
500 code, afs_inet_ntoa_r(ts->addr[0], hoststr));
502 ts->beaconSinceDown = 0;
503 urecovery_LostServer(ts);
505 ts->lastVoteTime = code;
506 if (code < oldestYesVote)
507 oldestYesVote = code;
512 yesVotes++; /* the extra epsilon */
513 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
514 ts->beaconSinceDown = 1;
515 ubik_dprint("yes vote from host %s\n",
516 afs_inet_ntoa_r(ts->addr[0], hoststr));
518 } else if (code == 0) {
519 ts->lastVoteTime = temp;
521 ts->beaconSinceDown = 1;
522 ubik_dprint("no vote from %s\n",
523 afs_inet_ntoa_r(ts->addr[0], hoststr));
524 } else if (code < 0) {
526 ts->beaconSinceDown = 0;
527 urecovery_LostServer(ts);
528 ubik_dprint("time out from %s\n",
529 afs_inet_ntoa_r(ts->addr[0], hoststr));
535 /* now call our own voter module to see if we'll vote for ourself. Note that
536 * the same restrictions apply for our voting for ourself as for our voting
537 * for anyone else. */
538 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
543 yesVotes++; /* extra epsilon */
544 if (i < oldestYesVote)
548 /* now decide if we have enough votes to become sync site.
549 * Note that we can still get enough votes even if we didn't for ourself. */
550 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
552 if (!beacon_globals.ubik_amSyncSite)
553 ubik_dprint("Ubik: I am the sync site\n");
554 beacon_globals.ubik_amSyncSite = 1;
555 beacon_globals.syncSiteUntil = oldestYesVote + SMALLTIME;
556 #ifndef AFS_PTHREAD_ENV
557 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
558 this may be a spurious signal call -- sjenkins */
559 LWP_NoYieldSignal(&beacon_globals.ubik_amSyncSite);
564 if (beacon_globals.ubik_amSyncSite)
565 ubik_dprint("Ubik: I am no longer the sync site\n");
566 beacon_globals.ubik_amSyncSite = 0;
569 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
578 * \brief Verify that a given IP addresses does actually exist on this machine.
580 * \param ame the pointer to my IP address specified in the
582 * \param aservers an array containing IP
583 * addresses of remote ubik servers. The array is
584 * terminated by a zero address.
586 * Algorithm : Verify that my IP addresses \p ame does actually exist
587 * on this machine. If any of my IP addresses are there
588 * in the remote server list \p aserver, remove them from
589 * this list. Update global variable \p ubik_host[] with
592 * \return 0 on success, non-zero on failure
595 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
596 afs_uint32 aservers[]) {
597 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
598 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
600 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
604 totalServers = info->numServers;
605 else { /* count the number of servers */
606 for (totalServers = 0, servList = aservers; *servList; servList++)
610 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
612 * Find addresses we are supposed to register as per the netrestrict file
613 * if it exists, else just register all the addresses we find on this
614 * host as returned by rx_getAllAddr (in NBO)
617 count = afsconf_ParseNetFiles(myAddr, NULL, NULL,
618 UBIK_MAX_INTERFACE_ADDR, reason,
619 AFSDIR_SERVER_NETINFO_FILEPATH,
620 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
622 ubik_print("ubik: Can't register any valid addresses:%s\n",
624 ubik_print("Aborting..\n");
629 /* get all my interface addresses in net byte order */
630 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
633 if (count <= 0) { /* no address found */
634 ubik_print("ubik: No network addresses found, aborting..");
638 /* verify that the My-address passed in by ubik is correct */
639 for (j = 0, found = 0; j < count; j++) {
640 if (*ame == myAddr[j]) { /* both in net byte order */
647 ubik_print("ubik: primary address %s does not exist\n",
648 afs_inet_ntoa_r(*ame, hoststr));
649 /* if we had the result of rx_getAllAddr already, avoid subverting
650 * the "is gethostbyname(gethostname()) us" check. If we're
651 * using NetInfo/NetRestrict, we assume they have enough clue
652 * to avoid that big hole in their foot from the loaded gun. */
654 /* take the address we did get, then see if ame was masked */
656 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
657 if (tcount <= 0) { /* no address found */
658 ubik_print("ubik: No network addresses found, aborting..");
662 /* verify that the My-address passed in by ubik is correct */
663 for (j = 0, found = 0; j < tcount; j++) {
664 if (*ame == myAddr2[j]) { /* both in net byte order */
674 /* if any of my addresses are there in serverList, then
675 ** use that as my primary addresses : the higher level
676 ** application screwed up in dealing with multihomed concepts
678 for (j = 0, found = 0; j < count; j++) {
679 for (i = 0; i < totalServers; i++) {
681 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
683 tmpAddr = aservers[i];
684 if (myAddr[j] == tmpAddr) {
693 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
696 /* get rid of servers which were purged because all
697 ** those interface addresses are myself
699 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
700 /* find the first zero entry from the beginning */
701 for (; (start < end) && (aservers[start]); start++);
703 /* find the last non-zero entry from the end */
704 for (; (end >= 0) && (!aservers[end]); end--);
706 /* if there is nothing more to purge, exit from loop */
711 aservers[start] = aservers[end];
712 aservers[end] = 0; /* this entry was moved */
716 /* update all my addresses in ubik_host in such a way
717 ** that ubik_host[0] has the primary address
720 for (j = 0, i = 1; j < count; j++)
721 if (*ame != myAddr[j])
722 ubik_host[i++] = myAddr[j];
724 return 0; /* return success */
729 * \brief Exchange IP address information with remote servers.
731 * \param ubik_host an array containing all my IP addresses.
733 * Algorithm : Do an RPC to all remote ubik servers informing them
734 * about my IP addresses. Get their IP addresses and
735 * update my linked list of ubik servers \p ubik_servers
737 * \return 0 on success, non-zero on failure
740 ubeacon_updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
742 int j, count, code = 0;
743 UbikInterfaceAddr inAddr, outAddr;
744 struct rx_connection *conns[MAXSERVERS];
745 struct ubik_server *ts, *server[MAXSERVERS];
750 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
751 conns[count] = ts->disk_rxcid;
757 /* inform all other servers only if there are more than one
758 * database servers in the cell */
762 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
763 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
766 /* do the multi-RX RPC to all other servers */
767 multi_Rx(conns, count) {
768 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
769 ts = server[multi_i]; /* reply received from this server */
772 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
774 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
775 ubik_print("ubik:Two primary addresses for same server \
777 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
779 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
780 ts->addr[j] = htonl(outAddr.hostAddr[j]);
783 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
786 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
787 afs_inet_ntoa_r(ts->addr[0], hoststr));
789 } else if (multi_error == UBADHOST) {
790 code = UBADHOST; /* remote CellServDB inconsistency */
791 ubik_print("Inconsistent Cell Info on server: ");
793 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
794 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
799 ts->up = 0; /* mark the remote server as down */
809 ubik_SetClientSecurityProcs(int (*secproc) (void *,
810 struct rx_securityClass **,
812 int (*checkproc) (void *),
815 secLayerProc = secproc;
816 tokenCheckProc = checkproc;