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 int nServers; /*!< total number of servers */
41 static char amIMagic = 0; /*!< is this host the magic host */
42 char amIClone = 0; /*!< is this a clone which doesn't vote */
43 static char ubik_singleServer = 0;
45 static int (*secLayerProc) (void *rock, struct rx_securityClass **,
47 static int (*tokenCheckProc) (void *rock) = NULL;
48 static void * securityRock = NULL;
50 afs_int32 ubikSecIndex;
51 struct rx_securityClass *ubikSecClass;
53 /* Values protected by the address lock */
54 struct addr_data addr_globals;
56 /* Values protected by the beacon lock */
57 struct beacon_data beacon_globals;
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 = beacon_globals.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 */
128 if (beacon_globals.ubik_amSyncSite == 0 || amIClone)
129 rcode = 0; /* if I don't think I'm the sync site, say so */
131 now = FT_ApproxTime();
132 if (beacon_globals.syncSiteUntil <= now) { /* if my votes have expired, say so */
133 if (beacon_globals.ubik_amSyncSite)
134 ubik_dprint("Ubik: I am no longer the sync site\n");
135 beacon_globals.ubik_amSyncSite = 0;
138 rcode = 1; /* otherwise still have the required votes */
142 urecovery_ResetState(); /* force recovery to re-execute */
144 ubik_dprint("beacon: amSyncSite is %d\n", rcode);
149 * \see ubeacon_InitServerListCommon()
152 ubeacon_InitServerListByInfo(afs_uint32 ame, struct afsconf_cell *info,
157 code = ubeacon_InitServerListCommon(ame, info, clones, 0);
162 * \param ame "address of me"
163 * \param aservers list of other servers
165 * \see ubeacon_InitServerListCommon()
168 ubeacon_InitServerList(afs_uint32 ame, afs_uint32 aservers[])
173 ubeacon_InitServerListCommon(ame, (struct afsconf_cell *)0, 0,
178 /* Must be called with address lock held */
180 ubeacon_InitSecurityClass(void)
183 /* get the security index to use, if we can */
185 i = (*secLayerProc) (securityRock, &addr_globals.ubikSecClass, &addr_globals.ubikSecIndex);
186 } else if (ubik_CRXSecurityProc) {
187 i = (*ubik_CRXSecurityProc) (ubik_CRXSecurityRock, &addr_globals.ubikSecClass,
188 &addr_globals.ubikSecIndex);
192 /* don't have sec module yet */
193 addr_globals.ubikSecIndex = 0;
194 addr_globals.ubikSecClass = rxnull_NewClientSecurityObject();
199 ubeacon_ReinitServer(struct ubik_server *ts)
201 if (tokenCheckProc && !(*tokenCheckProc) (securityRock)) {
202 struct rx_connection *disk_rxcid;
203 struct rx_connection *vote_rxcid;
204 struct rx_connection *tmp;
206 ubeacon_InitSecurityClass();
208 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->disk_rxcid)),
209 ubik_callPortal, DISK_SERVICE_ID,
210 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
212 tmp = ts->disk_rxcid;
213 ts->disk_rxcid = disk_rxcid;
214 rx_PutConnection(tmp);
217 rx_NewConnection(rx_HostOf(rx_PeerOf(ts->vote_rxcid)),
218 ubik_callPortal, VOTE_SERVICE_ID,
219 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
221 tmp = ts->vote_rxcid;
222 ts->vote_rxcid = vote_rxcid;
223 rx_PutConnection(tmp);
230 * \brief setup server list
232 * \param ame "address of me"
233 * \param aservers list of other servers
235 * called only at initialization to set up the list of servers to
236 * contact for votes. Just creates the server structure.
238 * The "magic" host is the one with the lowest internet address. It is
239 * magic because its vote counts epsilon more than the others. This acts
240 * as a tie-breaker when we have an even number of hosts in the system.
241 * For example, if the "magic" host is up in a 2 site system, then it
242 * is sync site. Without the magic host hack, if anyone crashed in a 2
243 * site system, we'd be out of business.
245 * \note There are two connections in every server structure, one for
246 * vote calls (which must always go through quickly) and one for database
247 * operations, which are subject to waiting for locks. If we used only
248 * one, the votes would sometimes get held up behind database operations,
249 * and the sync site guarantees would timeout even though the host would be
250 * up for communication.
252 * \see ubeacon_InitServerList(), ubeacon_InitServerListByInfo()
255 ubeacon_InitServerListCommon(afs_uint32 ame, struct afsconf_cell *info,
256 char clones[], afs_uint32 aservers[])
258 struct ubik_server *ts;
263 struct ubik_server *magicServer;
265 /* verify that the addresses passed in are correct */
266 if ((code = verifyInterfaceAddress(&ame, info, aservers)))
269 ubeacon_InitSecurityClass();
271 magicHost = ntohl(ame); /* do comparisons in host order */
272 magicServer = (struct ubik_server *)0;
275 for (i = 0; i < info->numServers; i++) {
276 if (ntohl((afs_uint32) info->hostAddr[i].sin_addr.s_addr) ==
277 ntohl((afs_uint32) ame)) {
286 for (i = 0; i < info->numServers; i++) {
289 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
290 memset(ts, 0, sizeof(struct ubik_server));
291 ts->next = ubik_servers;
293 ts->addr[0] = info->hostAddr[i].sin_addr.s_addr;
298 || ntohl((afs_uint32) ts->addr[0]) <
299 (afs_uint32) magicHost) {
300 magicHost = ntohl(ts->addr[0]);
307 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
308 ubik_callPortal, VOTE_SERVICE_ID,
309 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
312 rx_NewConnection(info->hostAddr[i].sin_addr.s_addr,
313 ubik_callPortal, DISK_SERVICE_ID,
314 addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
319 while ((servAddr = *aservers++)) {
321 return UNHOSTS; /* too many hosts */
322 ts = (struct ubik_server *)malloc(sizeof(struct ubik_server));
323 memset(ts, 0, sizeof(struct ubik_server));
324 ts->next = ubik_servers;
326 ts->addr[0] = servAddr; /* primary address in net byte order */
327 ts->vote_rxcid = rx_NewConnection(servAddr, ubik_callPortal, VOTE_SERVICE_ID,
328 addr_globals.ubikSecClass, addr_globals.ubikSecIndex); /* for vote reqs */
329 ts->disk_rxcid = rx_NewConnection(servAddr, ubik_callPortal, DISK_SERVICE_ID,
330 addr_globals.ubikSecClass, addr_globals.ubikSecIndex); /* for disk reqs */
331 ts->isClone = 0; /* don't know about clones */
333 if (ntohl((afs_uint32) servAddr) < (afs_uint32) magicHost) {
334 magicHost = ntohl(servAddr);
341 magicServer->magic = 1; /* remember for when counting votes */
343 if (!amIClone && !magicServer)
347 ++nServers; /* count this server as well as the remotes */
349 nServers = i + 1; /* count this server as well as the remotes */
351 ubik_quorum = (nServers >> 1) + 1; /* compute the majority figure */
352 /* send addrs to all other servers */
353 code = updateUbikNetworkAddress(ubik_host);
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;
401 /* loop forever getting votes */
402 lastWakeupTime = 0; /* keep track of time we last started a vote collection */
405 /* don't wakeup more than every POLLTIME seconds */
406 temp = (lastWakeupTime + POLLTIME) - FT_ApproxTime();
407 /* don't sleep if last collection phase took too long (probably timed someone out ) */
413 #ifdef AFS_PTHREAD_ENV
414 code = select(0, 0, 0, 0, &tt);
416 code = IOMGR_Select(0, 0, 0, 0, &tt);
421 lastWakeupTime = FT_ApproxTime(); /* started a new collection phase */
423 if (ubik_singleServer)
424 continue; /* special-case 1 server for speedy startup */
426 if (!uvote_ShouldIRun())
427 continue; /* if voter has heard from a better candidate than us, don't bother running */
429 /* otherwise we should run for election, or we're the sync site (and have already won);
430 * send out the beacon packets */
431 /* build list of all up hosts (noticing dead hosts are running again
432 * is a task for the recovery module, not the beacon module), and
433 * prepare to send them an r multi-call containing the beacon message */
434 i = 0; /* collect connections */
437 for (ts = ubik_servers; ts; ts = ts->next) {
438 if (ts->up && ts->addr[0] != ubik_host[0]) {
440 connections[i++] = ts->vote_rxcid;
445 servers[i] = (struct ubik_server *)0; /* end of list */
446 /* note that we assume in the vote module that we'll always get at least BIGTIME
447 * seconds of vote from anyone who votes for us, which means we can conservatively
448 * assume we'll be fine until SMALLTIME seconds after we start collecting votes */
449 /* this next is essentially an expansion of rgen's ServBeacon routine */
451 ttid.epoch = ubik_epochTime;
452 if (ubik_dbase->flags & DBWRITING) {
454 * if a write is in progress, we have to send the writeTidCounter
455 * which holds the tid counter of the write transaction , and not
456 * send the tidCounter value which holds the tid counter of the
459 ttid.counter = ubik_dbase->writeTidCounter;
461 ttid.counter = ubik_dbase->tidCounter + 1;
463 /* now analyze return codes, counting up our votes */
464 yesVotes = 0; /* count how many to ensure we have quorum */
465 oldestYesVote = 0x7fffffff; /* time quorum expires */
466 syncsite = ubeacon_AmSyncSite();
467 startTime = FT_ApproxTime();
469 * Don't waste time using mult Rx calls if there are no connections out there
473 multi_Rx(connections, i) {
474 multi_VOTE_Beacon(syncsite, startTime, &ubik_dbase->version,
476 temp = FT_ApproxTime(); /* now, more or less */
477 ts = servers[multi_i];
479 ts->lastBeaconSent = temp;
481 /* note that the vote time (the return code) represents the time
482 * the vote was computed, *not* the time the vote expires. We compute
483 * the latter down below if we got enough votes to go with */
485 if ((code & ~0xff) == ERROR_TABLE_BASE_RXK) {
486 ubik_dprint("token error %d from host %s\n",
487 code, afs_inet_ntoa_r(ts->addr[0], hoststr));
489 ts->beaconSinceDown = 0;
490 urecovery_LostServer(ts);
492 ts->lastVoteTime = code;
493 if (code < oldestYesVote)
494 oldestYesVote = code;
499 yesVotes++; /* the extra epsilon */
500 ts->up = 1; /* server is up (not really necessary: recovery does this for real) */
501 ts->beaconSinceDown = 1;
502 ubik_dprint("yes vote from host %s\n",
503 afs_inet_ntoa_r(ts->addr[0], hoststr));
505 } else if (code == 0) {
506 ts->lastVoteTime = temp;
508 ts->beaconSinceDown = 1;
509 ubik_dprint("no vote from %s\n",
510 afs_inet_ntoa_r(ts->addr[0], hoststr));
511 } else if (code < 0) {
513 ts->beaconSinceDown = 0;
514 urecovery_LostServer(ts);
515 ubik_dprint("time out from %s\n",
516 afs_inet_ntoa_r(ts->addr[0], hoststr));
522 /* now call our own voter module to see if we'll vote for ourself. Note that
523 * the same restrictions apply for our voting for ourself as for our voting
524 * for anyone else. */
525 i = SVOTE_Beacon((struct rx_call *)0, ubeacon_AmSyncSite(), startTime,
526 &ubik_dbase->version, &ttid);
530 yesVotes++; /* extra epsilon */
531 if (i < oldestYesVote)
535 /* now decide if we have enough votes to become sync site.
536 * Note that we can still get enough votes even if we didn't for ourself. */
537 if (yesVotes > nServers) { /* yesVotes is bumped by 2 or 3 for each site */
539 if (!beacon_globals.ubik_amSyncSite)
540 ubik_dprint("Ubik: I am the sync site\n");
541 beacon_globals.ubik_amSyncSite = 1;
542 beacon_globals.syncSiteUntil = oldestYesVote + SMALLTIME;
543 #ifndef AFS_PTHREAD_ENV
544 /* I did not find a corresponding LWP_WaitProcess(&ubik_amSyncSite) --
545 this may be a spurious signal call -- sjenkins */
546 LWP_NoYieldSignal(&beacon_globals.ubik_amSyncSite);
551 if (beacon_globals.ubik_amSyncSite)
552 ubik_dprint("Ubik: I am no longer the sync site\n");
553 beacon_globals.ubik_amSyncSite = 0;
555 urecovery_ResetState(); /* tell recovery we're no longer the sync site */
563 * \brief Verify that a given IP addresses does actually exist on this machine.
565 * \param ame the pointer to my IP address specified in the
567 * \param aservers an array containing IP
568 * addresses of remote ubik servers. The array is
569 * terminated by a zero address.
571 * Algorithm : Verify that my IP addresses \p ame does actually exist
572 * on this machine. If any of my IP addresses are there
573 * in the remote server list \p aserver, remove them from
574 * this list. Update global variable \p ubik_host[] with
577 * \return 0 on success, non-zero on failure
580 verifyInterfaceAddress(afs_uint32 *ame, struct afsconf_cell *info,
581 afs_uint32 aservers[]) {
582 afs_uint32 myAddr[UBIK_MAX_INTERFACE_ADDR], *servList, tmpAddr;
583 afs_uint32 myAddr2[UBIK_MAX_INTERFACE_ADDR];
585 int tcount, count, found, i, j, totalServers, start, end, usednetfiles =
589 totalServers = info->numServers;
590 else { /* count the number of servers */
591 for (totalServers = 0, servList = aservers; *servList; servList++)
595 if (AFSDIR_SERVER_NETRESTRICT_FILEPATH || AFSDIR_SERVER_NETINFO_FILEPATH) {
597 * Find addresses we are supposed to register as per the netrestrict file
598 * if it exists, else just register all the addresses we find on this
599 * host as returned by rx_getAllAddr (in NBO)
603 parseNetFiles(myAddr, NULL, NULL, UBIK_MAX_INTERFACE_ADDR, reason,
604 AFSDIR_SERVER_NETINFO_FILEPATH,
605 AFSDIR_SERVER_NETRESTRICT_FILEPATH);
607 ubik_print("ubik: Can't register any valid addresses:%s\n",
609 ubik_print("Aborting..\n");
614 /* get all my interface addresses in net byte order */
615 count = rx_getAllAddr(myAddr, UBIK_MAX_INTERFACE_ADDR);
618 if (count <= 0) { /* no address found */
619 ubik_print("ubik: No network addresses found, aborting..");
623 /* verify that the My-address passed in by ubik is correct */
624 for (j = 0, found = 0; j < count; j++) {
625 if (*ame == myAddr[j]) { /* both in net byte order */
632 ubik_print("ubik: primary address %s does not exist\n",
633 afs_inet_ntoa_r(*ame, hoststr));
634 /* if we had the result of rx_getAllAddr already, avoid subverting
635 * the "is gethostbyname(gethostname()) us" check. If we're
636 * using NetInfo/NetRestrict, we assume they have enough clue
637 * to avoid that big hole in their foot from the loaded gun. */
639 /* take the address we did get, then see if ame was masked */
641 tcount = rx_getAllAddr(myAddr2, UBIK_MAX_INTERFACE_ADDR);
642 if (tcount <= 0) { /* no address found */
643 ubik_print("ubik: No network addresses found, aborting..");
647 /* verify that the My-address passed in by ubik is correct */
648 for (j = 0, found = 0; j < tcount; j++) {
649 if (*ame == myAddr2[j]) { /* both in net byte order */
659 /* if any of my addresses are there in serverList, then
660 ** use that as my primary addresses : the higher level
661 ** application screwed up in dealing with multihomed concepts
663 for (j = 0, found = 0; j < count; j++) {
664 for (i = 0; i < totalServers; i++) {
666 tmpAddr = (afs_uint32) info->hostAddr[i].sin_addr.s_addr;
668 tmpAddr = aservers[i];
669 if (myAddr[j] == tmpAddr) {
678 ubik_print("Using %s as my primary address\n", afs_inet_ntoa_r(*ame, hoststr));
681 /* get rid of servers which were purged because all
682 ** those interface addresses are myself
684 for (start = 0, end = totalServers - 1; (start < end); start++, end--) {
685 /* find the first zero entry from the beginning */
686 for (; (start < end) && (aservers[start]); start++);
688 /* find the last non-zero entry from the end */
689 for (; (end >= 0) && (!aservers[end]); end--);
691 /* if there is nothing more to purge, exit from loop */
696 aservers[start] = aservers[end];
697 aservers[end] = 0; /* this entry was moved */
701 /* update all my addresses in ubik_host in such a way
702 ** that ubik_host[0] has the primary address
705 for (j = 0, i = 1; j < count; j++)
706 if (*ame != myAddr[j])
707 ubik_host[i++] = myAddr[j];
709 return 0; /* return success */
714 * \brief Exchange IP address information with remote servers.
716 * \param ubik_host an array containing all my IP addresses.
718 * Algorithm : Do an RPC to all remote ubik servers infroming them
719 * about my IP addresses. Get their IP addresses and
720 * update my linked list of ubik servers \p ubik_servers
722 * \return 0 on success, non-zero on failure
725 updateUbikNetworkAddress(afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR])
727 int j, count, code = 0;
728 UbikInterfaceAddr inAddr, outAddr;
729 struct rx_connection *conns[MAXSERVERS];
730 struct ubik_server *ts, *server[MAXSERVERS];
735 for (count = 0, ts = ubik_servers; ts; count++, ts = ts->next) {
736 conns[count] = ts->disk_rxcid;
742 /* inform all other servers only if there are more than one
743 * database servers in the cell */
747 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++)
748 inAddr.hostAddr[j] = ntohl(ubik_host[j]);
751 /* do the multi-RX RPC to all other servers */
752 multi_Rx(conns, count) {
753 multi_DISK_UpdateInterfaceAddr(&inAddr, &outAddr);
754 ts = server[multi_i]; /* reply received from this server */
757 if (ts->addr[0] != htonl(outAddr.hostAddr[0])) {
759 strcpy(buffer, afs_inet_ntoa_r(ts->addr[0], hoststr));
760 ubik_print("ubik:Two primary addresses for same server \
762 afs_inet_ntoa_r(htonl(outAddr.hostAddr[0]), hoststr));
764 for (j = 1; j < UBIK_MAX_INTERFACE_ADDR; j++)
765 ts->addr[j] = htonl(outAddr.hostAddr[j]);
768 } else if (multi_error == RXGEN_OPCODE) { /* pre 3.5 remote server */
771 ("ubik server %s does not support UpdateInterfaceAddr RPC\n",
772 afs_inet_ntoa_r(ts->addr[0], hoststr));
774 } else if (multi_error == UBADHOST) {
775 code = UBADHOST; /* remote CellServDB inconsistency */
776 ubik_print("Inconsistent Cell Info on server: ");
778 for (j = 0; j < UBIK_MAX_INTERFACE_ADDR && ts->addr[j]; j++)
779 ubik_print("%s ", afs_inet_ntoa_r(ts->addr[j], hoststr));
784 ts->up = 0; /* mark the remote server as down */
794 ubik_SetClientSecurityProcs(int (*secproc) (void *,
795 struct rx_securityClass **,
797 int (*checkproc) (void *),
800 secLayerProc = secproc;
801 tokenCheckProc = checkproc;