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>
23 #include <afs/netutils.h>
26 #define UBIK_INTERNALS
30 /* These global variables were used to set the function to use to initialise
31 * the client security layer. They are retained for backwards compatiblity with
32 * legacy callers - the ubik_SetClientSecurityProcs() interface should be used
35 int (*ubik_CRXSecurityProc) (void *rock, struct rx_securityClass **,
37 void *ubik_CRXSecurityRock;
39 /*! \name statics used to determine if we're the sync site */
40 static afs_int32 syncSiteUntil = 0; /*!< valid only if amSyncSite */
41 int ubik_amSyncSite = 0; /*!< flag telling if I'm sync site */
42 static int nServers; /*!< total number of servers */
43 static char amIMagic = 0; /*!< is this host the magic host */
44 char amIClone = 0; /*!< is this a clone which doesn't vote */
45 static char ubik_singleServer = 0;
47 static int (*secLayerProc) (void *rock, struct rx_securityClass **,
49 static int (*tokenCheckProc) (void *rock) = NULL;
50 static void * securityRock = NULL;
52 afs_int32 ubikSecIndex;
53 struct rx_securityClass *ubikSecClass;
54 static int ubeacon_InitServerListCommon(afs_uint32 ame,
55 struct afsconf_cell *info,
57 afs_uint32 aservers[]);
58 static int verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
59 afs_uint32 aservers[]);
60 static int updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR]);
64 * Module responsible for both deciding if we're currently the sync site,
65 * and keeping collecting votes so as to stay sync site.
67 * The basic module contacts all of the servers it can, trying to get them to vote
68 * for this server for sync site. The vote request message (called a beacon message)
69 * also specifies until which time this site claims to be the sync site, if at all, thus enabling
70 * receiving sites to know how long the sync site guarantee is made for.
72 * Each of these beacon messages is thus both a declaration of how long this site will
73 * remain sync site, and an attempt to extend that time by collecting votes for a later
74 * sync site extension.
76 * The voting module is responsible for choosing a reasonable time until which it promises
77 * not to vote for someone else. This parameter (BIG seconds) is not actually passed in
78 * the interface (perhaps it should be?) but is instead a compile time constant that both
81 * The beacon and vote modules work intimately together; the vote module decides how long
82 * it should promise the beacon module its vote, and the beacon module takes all of these
83 * votes and decides for how long it is the synchronization site.
86 /*! \brief procedure called from debug rpc call to get this module's state for debugging */
88 ubeacon_Debug(struct ubik_debug *aparm)
90 /* fill in beacon's state fields in the ubik_debug structure */
91 aparm->syncSiteUntil = syncSiteUntil;
92 aparm->nServers = nServers;
96 * \brief Procedure that determines whether this site has enough current votes to remain sync site.
98 * Called from higher-level modules (everything but the vote module).
100 * If we're the sync site, check that our guarantees, obtained by the ubeacon_Interact()
101 * light-weight process, haven't expired. We're sync site as long as a majority of the
102 * servers in existence have promised us unexpired guarantees. The variable #ubik_syncSiteUntil
103 * contains the time at which the latest of the majority of the sync site guarantees expires
104 * (if the variable #ubik_amSyncSite is true)
105 * This module also calls up to the recovery module if it thinks that the recovery module
106 * may have to pick up a new database (which offucr sif [sic] we lose the sync site votes).
108 * \return 1 if local site is the sync site
109 * \return 0 if sync site is elsewhere
112 ubeacon_AmSyncSite(void)
117 /* special case for fast startup */
118 if (nServers == 1 && !amIClone) {
119 return 1; /* one guy is always the sync site */
122 if (ubik_amSyncSite == 0 || amIClone)
123 rcode = 0; /* if I don't think I'm the sync site, say so */
125 now = FT_ApproxTime();
126 if (syncSiteUntil <= now) { /* if my votes have expired, say so */
128 ubik_dprint("Ubik: I am no longer the sync site\n");
132 rcode = 1; /* otherwise still have the required votes */
136 urecovery_ResetState(); /* force recovery to re-execute */
137 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
142 * \see ubeacon_InitServerListCommon()
145 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
150 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
155 * \param ame "address of me"
156 * \param aservers list of other servers
158 * \see ubeacon_InitServerListCommon()
161 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
166 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
172 ubeacon_InitSecurityClass(void)
175 /* get the security index to use, if we can */
177 i = (*secLayerProc) (securityRock, &ubikSecClass, &ubikSecIndex);
178 } else if (ubik_CRXSecurityProc) {
179 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &ubikSecClass,
184 /* don't have sec module yet */
186 ubikSecClass = rxnull_NewClientSecurityObject();
191 ubeacon_ReinitServer(struct ubik_server *ts)
193 if (tokenCheckProc && !(*tokenCheckProc) (securityRock)) {
194 struct rx_connection *disk_rxcid;
195 struct rx_connection *vote_rxcid;
196 struct rx_connection *tmp;
197 ubeacon_InitSecurityClass();
199 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->disk_rxcid)),
200 ubik_callPortal, DISK_SERVICE_ID,
201 ubikSecClass, ubikSecIndex);
203 tmp = ts->disk_rxcid;
204 ts->disk_rxcid = disk_rxcid;
205 rx_PutConnection(tmp);
208 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->vote_rxcid)),
209 ubik_callPortal, VOTE_SERVICE_ID,
210 ubikSecClass, ubikSecIndex);
212 tmp = ts->vote_rxcid;
213 ts->vote_rxcid = vote_rxcid;
214 rx_PutConnection(tmp);
220 * \brief setup server list
222 * \param ame "address of me"
223 * \param aservers list of other servers
225 * called only at initialization to set up the list of servers to
226 * contact for votes. Just creates the server structure.
228 * The "magic" host is the one with the lowest internet address. It is
229 * magic because its vote counts epsilon more than the others. This acts
230 * as a tie-breaker when we have an even number of hosts in the system.
231 * For example, if the "magic" host is up in a 2 site system, then it
232 * is sync site. Without the magic host hack, if anyone crashed in a 2
233 * site system, we'd be out of business.
235 * \note There are two connections in every server structure, one for
236 * vote calls (which must always go through quickly) and one for database
237 * operations, which are subject to waiting for locks. If we used only
238 * one, the votes would sometimes get held up behind database operations,
239 * and the sync site guarantees would timeout even though the host would be
240 * up for communication.
242 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
245 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
246 char clones[], afs_uint32 aservers[])
248 struct ubik_server *ts;
253 struct ubik_server *magicServer;
255 /* verify that the addresses passed in are correct */
256 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
259 ubeacon_InitSecurityClass();
261 magicHost = ntohl(ame); /* do comparisons in host order */
262 magicServer = (struct ubik_server *)0;
265 for (i = 0; i < info->numServers; i++) {
266 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
267 ntohl((afs_uint32) ame)) {
276 for (i = 0; i < info->numServers; i++) {
279 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
280 memset(ts, 0, sizeof(struct ubik_server));
281 ts->next = ubik_servers;
283 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
288 || ntohl((afs_uint32) ts->addr[0]) <
289 (afs_uint32) magicHost) {
290 magicHost = ntohl(ts->addr[0]);
297 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
298 ubik_callPortal, VOTE_SERVICE_ID,
299 ubikSecClass, ubikSecIndex);
302 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
303 ubik_callPortal, DISK_SERVICE_ID,
304 ubikSecClass, ubikSecIndex);
309 while ((servAddr = *aservers++)) {
311 return UNHOSTS; /* too many hosts */
312 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
313 memset(ts, 0, sizeof(struct ubik_server));
314 ts->next = ubik_servers;
316 ts->addr[0] = servAddr; /* primary address in net byte order */
317 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for vote reqs */
318 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID, ubikSecClass, ubikSecIndex); /* for disk reqs */
319 ts->isClone = 0; /* don't know about clones */
321 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
322 magicHost = ntohl(servAddr);
329 magicServer->magic = 1; /* remember for when counting votes */
331 if (!amIClone && !magicServer)
335 ++nServers; /* count this server as well as the remotes */
337 nServers = i + 1; /* count this server as well as the remotes */
339 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
340 /* send addrs to all other servers */
341 code = updateUbikNetworkAddress(ubik_host);
345 /* Shoud we set some defaults for RX??
347 r_nRetries = (RPCTIMEOUT/r_retryInterval);
350 if (!ubik_servers) /* special case 1 server */
351 ubik_singleServer = 1;
352 if (nServers == 1 && !amIClone) {
353 ubik_amSyncSite = 1; /* let's start as sync site */
354 syncSiteUntil = 0x7fffffff; /* and be it quite a while */
357 if (nServers == 1) /* special case 1 server */
358 ubik_singleServer = 1;
361 if (ubik_singleServer) {
362 if (!ubik_amSyncSite)
363 ubik_dprint("Ubik: I am the sync site - 1 server\n");
365 syncSiteUntil = 0x7fffffff; /* quite a while */
371 * \brief main lwp loop for code that sends out beacons.
373 * This code only runs while we're sync site or we want to be the sync site.
374 * It runs in its very own light-weight process.
377 ubeacon_Interact(void *dummy)
381 struct rx_connection *connections[MAXSERVERS];
382 struct ubik_server *servers[MAXSERVERS];
384 struct ubik_server *ts;
385 afs_int32 temp, yesVotes, lastWakeupTime, oldestYesVote, syncsite;
386 struct ubik_tid ttid;
389 /* loop forever getting votes */
390 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
393 /* don't wakeup more than every POLLTIME seconds */
394 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
395 /* don't sleep if last collection phase took too long (probably timed someone out ) */
401 #ifdef AFS_PTHREAD_ENV
402 code = select(0, 0, 0, 0, &tt);
404 code = IOMGR_Select(0, 0, 0, 0, &tt);
409 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
411 if (ubik_singleServer)
412 continue; /* special-case 1 server for speedy startup */
414 if (!uvote_ShouldIRun())
415 continue; /* if voter has heard from a better candidate than us, don't bother running */
417 /* otherwise we should run for election, or we're the sync site (and have already won);
418 * send out the beacon packets */
419 /* build list of all up hosts (noticing dead hosts are running again
420 * is a task for the recovery module, not the beacon module), and
421 * prepare to send them an r multi-call containing the beacon message */
422 i = 0; /* collect connections */
423 for (ts = ubik_servers; ts; ts = ts->next) {
424 if (ts->up && ts->addr[0] != ubik_host[0]) {
426 connections[i++] = ts->vote_rxcid;
429 servers[i] = (struct ubik_server *)0; /* end of list */
430 /* note that we assume in the vote module that we'll always get at least BIGTIME
431 * seconds of vote from anyone who votes for us, which means we can conservatively
432 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
433 /* this next is essentially an expansion of rgen's ServBeacon routine */
435 ttid.epoch = ubik_epochTime;
436 if (ubik_dbase->flags & DBWRITING) {
438 * if a write is in progress, we have to send the writeTidCounter
439 * which holds the tid counter of the write transaction , and not
440 * send the tidCounter value which holds the tid counter of the
443 ttid.counter = ubik_dbase->writeTidCounter;
445 ttid.counter = ubik_dbase->tidCounter + 1;
447 /* now analyze return codes, counting up our votes */
448 yesVotes = 0; /* count how many to ensure we have quorum */
449 oldestYesVote = 0x7fffffff; /* time quorum expires */
450 syncsite = ubeacon_AmSyncSite();
451 startTime = FT_ApproxTime();
453 * Don't waste time using mult Rx calls if there are no connections out there
457 multi_Rx(connections, i) {
458 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
460 temp = FT_ApproxTime(); /* now, more or less */
461 ts = servers[multi_i];
462 ts->lastBeaconSent = temp;
464 /* note that the vote time (the return code) represents the time
465 * the vote was computed, *not* the time the vote expires. We compute
466 * the latter down below if we got enough votes to go with */
468 if ((code & ~0xff) == ERROR_TABLE_BASE_RXK) {
469 ubik_dprint("token error %d from host %s\n",
470 code, afs_inet_ntoa_r(ts->addr[0], hoststr));
472 ts->beaconSinceDown = 0;
473 urecovery_LostServer(ts);
475 ts->lastVoteTime = code;
476 if (code < oldestYesVote)
477 oldestYesVote = code;
482 yesVotes++; /* the extra epsilon */
483 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
484 ts->beaconSinceDown = 1;
485 ubik_dprint("yes vote from host %s\n",
486 afs_inet_ntoa_r(ts->addr[0], hoststr));
488 } else if (code == 0) {
489 ts->lastVoteTime = temp;
491 ts->beaconSinceDown = 1;
492 ubik_dprint("no vote from %s\n",
493 afs_inet_ntoa_r(ts->addr[0], hoststr));
494 } else if (code < 0) {
496 ts->beaconSinceDown = 0;
497 urecovery_LostServer(ts);
498 ubik_dprint("time out from %s\n",
499 afs_inet_ntoa_r(ts->addr[0], hoststr));
504 /* now call our own voter module to see if we'll vote for ourself. Note that
505 * the same restrictions apply for our voting for ourself as for our voting
506 * for anyone else. */
507 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
508 &ubik_dbase->version, &ttid);
512 yesVotes++; /* extra epsilon */
513 if (i < oldestYesVote)
517 /* now decide if we have enough votes to become sync site.
518 * Note that we can still get enough votes even if we didn't for ourself. */
519 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
520 if (!ubik_amSyncSite)
521 ubik_dprint("Ubik: I am the sync site\n");
523 syncSiteUntil = oldestYesVote + SMALLTIME;
524 #ifndef AFS_PTHREAD_ENV
525 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
526 this may be a spurious signal call -- sjenkins */
527 LWP_NoYieldSignal(&ubik_amSyncSite);
531 ubik_dprint("Ubik: I am no longer the sync site\n");
533 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
541 * \brief Verify that a given IP addresses does actually exist on this machine.
543 * \param ame the pointer to my IP address specified in the
545 * \param aservers an array containing IP
546 * addresses of remote ubik servers. The array is
547 * terminated by a zero address.
549 * Algorithm : Verify that my IP addresses \p ame does actually exist
550 * on this machine. If any of my IP addresses are there
551 * in the remote server list \p aserver, remove them from
552 * this list. Update global variable \p ubik_host[] with
555 * \return 0 on success, non-zero on failure
558 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
559 afs_uint32 aservers[]) {
560 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
561 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
563 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
567 totalServers = info->numServers;
568 else { /* count the number of servers */
569 for (totalServers = 0, servList = aservers; *servList; servList++)
573 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
575 * Find addresses we are supposed to register as per the netrestrict file
576 * if it exists, else just register all the addresses we find on this
577 * host as returned by rx_getAllAddr (in NBO)
581 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
582 AFSDIR_SERVER_NETINFO_FILEPATH,
583 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
585 ubik_print("ubik: Can't register any valid addresses:%s\n",
587 ubik_print("Aborting..\n");
592 /* get all my interface addresses in net byte order */
593 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
596 if (count <= 0) { /* no address found */
597 ubik_print("ubik: No network addresses found, aborting..");
601 /* verify that the My-address passed in by ubik is correct */
602 for (j = 0, found = 0; j < count; j++) {
603 if (*ame == myAddr[j]) { /* both in net byte order */
610 ubik_print("ubik: primary address %s does not exist\n",
611 afs_inet_ntoa_r(*ame, hoststr));
612 /* if we had the result of rx_getAllAddr already, avoid subverting
613 * the "is gethostbyname(gethostname()) us" check. If we're
614 * using NetInfo/NetRestrict, we assume they have enough clue
615 * to avoid that big hole in their foot from the loaded gun. */
617 /* take the address we did get, then see if ame was masked */
619 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
620 if (tcount <= 0) { /* no address found */
621 ubik_print("ubik: No network addresses found, aborting..");
625 /* verify that the My-address passed in by ubik is correct */
626 for (j = 0, found = 0; j < tcount; j++) {
627 if (*ame == myAddr2[j]) { /* both in net byte order */
637 /* if any of my addresses are there in serverList, then
638 ** use that as my primary addresses : the higher level
639 ** application screwed up in dealing with multihomed concepts
641 for (j = 0, found = 0; j < count; j++) {
642 for (i = 0; i < totalServers; i++) {
644 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
646 tmpAddr = aservers[i];
647 if (myAddr[j] == tmpAddr) {
656 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
659 /* get rid of servers which were purged because all
660 ** those interface addresses are myself
662 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
663 /* find the first zero entry from the beginning */
664 for (; (start < end) && (aservers[start]); start++);
666 /* find the last non-zero entry from the end */
667 for (; (end >= 0) && (!aservers[end]); end--);
669 /* if there is nothing more to purge, exit from loop */
674 aservers[start] = aservers[end];
675 aservers[end] = 0; /* this entry was moved */
679 /* update all my addresses in ubik_host in such a way
680 ** that ubik_host[0] has the primary address
683 for (j = 0, i = 1; j < count; j++)
684 if (*ame != myAddr[j])
685 ubik_host[i++] = myAddr[j];
687 return 0; /* return success */
692 * \brief Exchange IP address information with remote servers.
694 * \param ubik_host an array containing all my IP addresses.
696 * Algorithm : Do an RPC to all remote ubik servers infroming them
697 * about my IP addresses. Get their IP addresses and
698 * update my linked list of ubik servers \p ubik_servers
700 * \return 0 on success, non-zero on failure
703 updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
705 int j, count, code = 0;
706 UbikInterfaceAddr inAddr, outAddr;
707 struct rx_connection *conns[MAXSERVERS];
708 struct ubik_server *ts, *server[MAXSERVERS];
712 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
713 conns[count] = ts->disk_rxcid;
718 /* inform all other servers only if there are more than one
719 * database servers in the cell */
723 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
724 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
727 /* do the multi-RX RPC to all other servers */
728 multi_Rx(conns, count) {
729 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
730 ts = server[multi_i]; /* reply received from this server */
732 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
734 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
735 ubik_print("ubik:Two primary addresses for same server \
737 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
739 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
740 ts->addr[j] = htonl(outAddr.hostAddr[j]);
742 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
744 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
745 afs_inet_ntoa_r(ts->addr[0], hoststr));
746 } else if (multi_error == UBADHOST) {
747 code = UBADHOST; /* remote CellServDB inconsistency */
748 ubik_print("Inconsistent Cell Info on server: ");
749 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
750 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
753 ts->up = 0; /* mark the remote server as down */
762 ubik_SetClientSecurityProcs(int (*secproc) (void *,
763 struct rx_securityClass **,
765 int (*checkproc) (void *),
768 secLayerProc = secproc;
769 tokenCheckProc = checkproc;