/* * Copyright 2000, International Business Machines Corporation and others. * All Rights Reserved. * * This software has been released under the terms of the IBM Public * License. For details, see the LICENSE file in the top-level source * directory or online at http://www.openafs.org/dl/license10.html */ #include #include #include #include #ifdef AFS_PTHREAD_ENV # include #else # include #endif #include #include #include #include #define UBIK_INTERNALS #include "ubik.h" #include "ubik_int.h" #include /* temporary hack by klm */ #define ERROR_EXIT(code) do { \ error = (code); \ goto error_exit; \ } while (0) /*! * \file * This system is organized in a hierarchical set of related modules. Modules * at one level can only call modules at the same level or below. * * At the bottom level (0) we have R, RFTP, LWP and IOMGR, i.e. the basic * operating system primitives. * * At the next level (1) we have * * \li VOTER--The module responsible for casting votes when asked. It is also * responsible for determining whether this server should try to become * a synchronization site. * \li BEACONER--The module responsible for sending keep-alives out when a * server is actually the sync site, or trying to become a sync site. * \li DISK--The module responsible for representing atomic transactions * on the local disk. It maintains a new-value only log. * \li LOCK--The module responsible for locking byte ranges in the database file. * * At the next level (2) we have * * \li RECOVERY--The module responsible for ensuring that all members of a quorum * have the same up-to-date database after a new synchronization site is * elected. This module runs only on the synchronization site. * * At the next level (3) we have * * \li REMOTE--The module responsible for interpreting requests from the sync * site and applying them to the database, after obtaining the appropriate * locks. * * At the next level (4) we have * * \li UBIK--The module users call to perform operations on the database. */ /* some globals */ afs_int32 ubik_quorum = 0; struct ubik_dbase *ubik_dbase = 0; struct ubik_stats ubik_stats; afs_uint32 ubik_host[UBIK_MAX_INTERFACE_ADDR]; afs_int32 urecovery_state = 0; int (*ubik_SyncWriterCacheProc) (void); struct ubik_server *ubik_servers; short ubik_callPortal; /* These global variables were used to control the server security layers. * They are retained for backwards compatibility with legacy callers. * * The ubik_SetServerSecurityProcs() interface should be used instead. */ int (*ubik_SRXSecurityProc) (void *, struct rx_securityClass **, afs_int32 *); void *ubik_SRXSecurityRock; int (*ubik_CheckRXSecurityProc) (void *, struct rx_call *); void *ubik_CheckRXSecurityRock; static int BeginTrans(struct ubik_dbase *dbase, afs_int32 transMode, struct ubik_trans **transPtr, int readAny); static struct rx_securityClass **ubik_sc = NULL; static void (*buildSecClassesProc)(void *, struct rx_securityClass ***, afs_int32 *) = NULL; static int (*checkSecurityProc)(void *, struct rx_call *) = NULL; static void *securityRock = NULL; struct version_data version_globals; #define CStampVersion 1 /* meaning set ts->version */ #define CCheckSyncAdvertised 2 /* check if the remote knows we are the sync-site */ static_inline struct rx_connection * Quorum_StartIO(struct ubik_trans *atrans, struct ubik_server *as) { struct rx_connection *conn; UBIK_ADDR_LOCK; conn = as->disk_rxcid; #ifdef AFS_PTHREAD_ENV rx_GetConnection(conn); UBIK_ADDR_UNLOCK; DBRELE(atrans->dbase); #else UBIK_ADDR_UNLOCK; #endif /* AFS_PTHREAD_ENV */ return conn; } static_inline void Quorum_EndIO(struct ubik_trans *atrans, struct rx_connection *aconn) { #ifdef AFS_PTHREAD_ENV DBHOLD(atrans->dbase); rx_PutConnection(aconn); #endif /* AFS_PTHREAD_ENV */ } /* * Iterate over all servers. Callers pass in *ts which is used to track * the current server. * - Returns 1 if there are no more servers * - Returns 0 with conn set to the connection for the current server if * it's up and current */ static int ContactQuorum_iterate(struct ubik_trans *atrans, int aflags, struct ubik_server **ts, struct rx_connection **conn, afs_int32 *rcode, afs_int32 *okcalls, afs_int32 code, const char *procname) { if (!*ts) { /* Initial call - start iterating over servers */ *ts = ubik_servers; *conn = NULL; *rcode = 0; *okcalls = 0; } else { if (*conn) { Quorum_EndIO(atrans, *conn); *conn = NULL; if (code) { /* failure */ char hoststr[16]; *rcode = code; UBIK_BEACON_LOCK; (*ts)->up = 0; /* mark as down now; beacons will no longer be sent */ (*ts)->beaconSinceDown = 0; UBIK_BEACON_UNLOCK; (*ts)->currentDB = 0; urecovery_LostServer(*ts); /* tell recovery to try to resend dbase later */ ViceLog(0, ("Server %s is marked down due to %s code %d\n", afs_inet_ntoa_r((*ts)->addr[0], hoststr), procname, *rcode)); } else { /* success */ if (!(*ts)->isClone) (*okcalls)++; /* count up how many worked */ if (aflags & CStampVersion) { (*ts)->version = atrans->dbase->version; } } } *ts = (*ts)->next; } if (!(*ts)) return 1; UBIK_BEACON_LOCK; if (!(*ts)->up || !(*ts)->currentDB || /* do not call DISK_Begin until we know that lastYesState is set on the * remote in question; otherwise, DISK_Begin will fail. */ ((aflags & CCheckSyncAdvertised) && !((*ts)->beaconSinceDown && (*ts)->lastVote))) { UBIK_BEACON_UNLOCK; (*ts)->currentDB = 0; /* db is no longer current; we just missed an update */ return 0; /* not up-to-date, don't bother. NULL conn will tell caller not to use */ } UBIK_BEACON_UNLOCK; *conn = Quorum_StartIO(atrans, *ts); return 0; } static int ContactQuorum_rcode(int okcalls, afs_int32 rcode) { /* * return 0 if we successfully contacted a quorum, otherwise return error code. * We don't have to contact ourselves (that was done locally) */ if (okcalls + 1 >= ubik_quorum) return 0; else return (rcode != 0) ? rcode : UNOQUORUM; } /*! * \brief Perform an operation at a quorum, handling error conditions. * \return 0 if all worked and a quorum was contacted successfully * \return otherwise mark failing server as down and return #UERROR * * \note If any server misses an update, we must wait #BIGTIME seconds before * allowing the transaction to commit, to ensure that the missing and * possibly still functioning server times out and stops handing out old * data. This is done in the commit code, where we wait for a server marked * down to have stayed down for #BIGTIME seconds before we allow a transaction * to commit. A server that fails but comes back up won't give out old data * because it is sent the sync count along with the beacon message that * marks it as \b really up (\p beaconSinceDown). */ static afs_int32 ContactQuorum_NoArguments(afs_int32 (*proc)(struct rx_connection *, ubik_tid *), struct ubik_trans *atrans, int aflags, const char *procname) { struct ubik_server *ts = NULL; afs_int32 code = 0, rcode, okcalls; struct rx_connection *conn; int done; done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); while (!done) { if (conn) code = (*proc)(conn, &atrans->tid); done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); } return ContactQuorum_rcode(okcalls, rcode); } static afs_int32 ContactQuorum_DISK_Lock(struct ubik_trans *atrans, int aflags,afs_int32 file, afs_int32 position, afs_int32 length, afs_int32 type) { struct ubik_server *ts = NULL; afs_int32 code = 0, rcode, okcalls; struct rx_connection *conn; int done; char *procname = "DISK_Lock"; done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); while (!done) { if (conn) code = DISK_Lock(conn, &atrans->tid, file, position, length, type); done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); } return ContactQuorum_rcode(okcalls, rcode); } static afs_int32 ContactQuorum_DISK_Truncate(struct ubik_trans *atrans, int aflags, afs_int32 file, afs_int32 length) { struct ubik_server *ts = NULL; afs_int32 code = 0, rcode, okcalls; struct rx_connection *conn; int done; char *procname = "DISK_Truncate"; done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); while (!done) { if (conn) code = DISK_Truncate(conn, &atrans->tid, file, length); done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); } return ContactQuorum_rcode(okcalls, rcode); } static afs_int32 ContactQuorum_DISK_WriteV(struct ubik_trans *atrans, int aflags, iovec_wrt * io_vector, iovec_buf *io_buffer) { struct ubik_server *ts = NULL; afs_int32 code = 0, rcode, okcalls; struct rx_connection *conn; int done; char *procname = "DISK_WriteV"; done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); while (!done) { if (conn) { procname = "DISK_WriteV"; /* in case previous fallback to DISK_Write */ code = DISK_WriteV(conn, &atrans->tid, io_vector, io_buffer); if ((code <= -450) && (code > -500)) { /* An RPC interface mismatch (as defined in comerr/error_msg.c). * Un-bulk the entries and do individual DISK_Write calls * instead of DISK_WriteV. */ struct ubik_iovec *iovec = (struct ubik_iovec *)io_vector->iovec_wrt_val; char *iobuf = (char *)io_buffer->iovec_buf_val; bulkdata tcbs; afs_int32 i, offset; procname = "DISK_Write"; /* for accurate error msg, if any */ for (i = 0, offset = 0; i < io_vector->iovec_wrt_len; i++) { /* Sanity check for going off end of buffer */ if ((offset + iovec[i].length) > io_buffer->iovec_buf_len) { code = UINTERNAL; break; } tcbs.bulkdata_len = iovec[i].length; tcbs.bulkdata_val = &iobuf[offset]; code = DISK_Write(conn, &atrans->tid, iovec[i].file, iovec[i].position, &tcbs); if (code) break; offset += iovec[i].length; } } } done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); } return ContactQuorum_rcode(okcalls, rcode); } afs_int32 ContactQuorum_DISK_SetVersion(struct ubik_trans *atrans, int aflags, ubik_version *OldVersion, ubik_version *NewVersion) { struct ubik_server *ts = NULL; afs_int32 code = 0, rcode, okcalls; struct rx_connection *conn; int done; char *procname = "DISK_SetVersion"; done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); while (!done) { if (conn) code = DISK_SetVersion(conn, &atrans->tid, OldVersion, NewVersion); done = ContactQuorum_iterate(atrans, aflags, &ts, &conn, &rcode, &okcalls, code, procname); } return ContactQuorum_rcode(okcalls, rcode); } #if defined(AFS_PTHREAD_ENV) static int ubik_thread_create(pthread_attr_t *tattr, pthread_t *thread, void *proc) { opr_Verify(pthread_attr_init(tattr) == 0); opr_Verify(pthread_attr_setdetachstate(tattr, PTHREAD_CREATE_DETACHED) == 0); opr_Verify(pthread_create(thread, tattr, proc, NULL) == 0); return 0; } #endif /*! * \brief This routine initializes the ubik system for a set of servers. * \return 0 for success, or an error code on failure. * \param serverList set of servers specified; nServers gives the number of entries in this array. * \param pathName provides an initial prefix used for naming storage files used by this system. * \param dbase the returned structure representing this instance of an ubik; it is passed to various calls below. * * \todo This routine should perhaps be generalized to a low-level disk interface providing read, write, file enumeration and sync operations. * * \warning The host named by myHost should not also be listed in serverList. * * \see ubik_ServerInit(), ubik_ServerInitByInfo() */ static int ubik_ServerInitCommon(afs_uint32 myHost, short myPort, struct afsconf_cell *info, char clones[], afs_uint32 serverList[], const char *pathName, struct ubik_dbase **dbase) { struct ubik_dbase *tdb; afs_int32 code; #ifdef AFS_PTHREAD_ENV pthread_t rxServerThread; /* pthread variables */ pthread_t ubeacon_InteractThread; pthread_t urecovery_InteractThread; pthread_attr_t rxServer_tattr; pthread_attr_t ubeacon_Interact_tattr; pthread_attr_t urecovery_Interact_tattr; #else PROCESS junk; extern int rx_stackSize; #endif afs_int32 secIndex; struct rx_securityClass *secClass; int numClasses; struct rx_service *tservice; initialize_U_error_table(); tdb = malloc(sizeof(struct ubik_dbase)); tdb->pathName = strdup(pathName); tdb->activeTrans = (struct ubik_trans *)0; memset(&tdb->version, 0, sizeof(struct ubik_version)); memset(&tdb->cachedVersion, 0, sizeof(struct ubik_version)); #ifdef AFS_PTHREAD_ENV opr_mutex_init(&tdb->versionLock); opr_mutex_init(&beacon_globals.beacon_lock); opr_mutex_init(&vote_globals.vote_lock); opr_mutex_init(&addr_globals.addr_lock); opr_mutex_init(&version_globals.version_lock); #else Lock_Init(&tdb->versionLock); #endif Lock_Init(&tdb->cache_lock); tdb->flags = 0; tdb->read = uphys_read; tdb->write = uphys_write; tdb->truncate = uphys_truncate; tdb->open = uphys_invalidate; /* this function isn't used any more */ tdb->sync = uphys_sync; tdb->stat = uphys_stat; tdb->getlabel = uphys_getlabel; tdb->setlabel = uphys_setlabel; tdb->getnfiles = uphys_getnfiles; tdb->buffered_append = uphys_buf_append; tdb->readers = 0; tdb->tidCounter = tdb->writeTidCounter = 0; *dbase = tdb; ubik_dbase = tdb; /* for now, only one db per server; can fix later when we have names for the other dbases */ #ifdef AFS_PTHREAD_ENV opr_cv_init(&tdb->version_cond); opr_cv_init(&tdb->flags_cond); #endif /* AFS_PTHREAD_ENV */ /* initialize RX */ /* the following call is idempotent so when/if it got called earlier, * by whatever called us, it doesn't really matter -- klm */ code = rx_Init(myPort); if (code < 0) return code; ubik_callPortal = myPort; udisk_Init(ubik_nBuffers); ulock_Init(); code = uvote_Init(); if (code) return code; code = urecovery_Initialize(tdb); if (code) return code; if (info) code = ubeacon_InitServerListByInfo(myHost, info, clones); else code = ubeacon_InitServerList(myHost, serverList); if (code) return code; /* try to get an additional security object */ if (buildSecClassesProc == NULL) { numClasses = 3; ubik_sc = calloc(numClasses, sizeof(struct rx_securityClass *)); ubik_sc[0] = rxnull_NewServerSecurityObject(); if (ubik_SRXSecurityProc) { code = (*ubik_SRXSecurityProc) (ubik_SRXSecurityRock, &secClass, &secIndex); if (code == 0) { ubik_sc[secIndex] = secClass; } } } else { (*buildSecClassesProc) (securityRock, &ubik_sc, &numClasses); } /* for backwards compat this should keep working as it does now and not host bind */ tservice = rx_NewService(0, VOTE_SERVICE_ID, "VOTE", ubik_sc, numClasses, VOTE_ExecuteRequest); if (tservice == (struct rx_service *)0) { ViceLog(0, ("Could not create VOTE rx service!\n")); return -1; } rx_SetMinProcs(tservice, 2); rx_SetMaxProcs(tservice, 3); tservice = rx_NewService(0, DISK_SERVICE_ID, "DISK", ubik_sc, numClasses, DISK_ExecuteRequest); if (tservice == (struct rx_service *)0) { ViceLog(0, ("Could not create DISK rx service!\n")); return -1; } rx_SetMinProcs(tservice, 2); rx_SetMaxProcs(tservice, 3); /* start an rx_ServerProc to handle incoming RPC's in particular the * UpdateInterfaceAddr RPC that occurs in ubeacon_InitServerList. This avoids * the "steplock" problem in ubik initialization. Defect 11037. */ #ifdef AFS_PTHREAD_ENV ubik_thread_create(&rxServer_tattr, &rxServerThread, (void *)rx_ServerProc); #else LWP_CreateProcess(rx_ServerProc, rx_stackSize, RX_PROCESS_PRIORITY, NULL, "rx_ServerProc", &junk); #endif /* send addrs to all other servers */ code = ubeacon_updateUbikNetworkAddress(ubik_host); if (code) return code; /* now start up async processes */ #ifdef AFS_PTHREAD_ENV ubik_thread_create(&ubeacon_Interact_tattr, &ubeacon_InteractThread, (void *)ubeacon_Interact); #else code = LWP_CreateProcess(ubeacon_Interact, 16384 /*8192 */ , LWP_MAX_PRIORITY - 1, (void *)0, "beacon", &junk); if (code) return code; #endif #ifdef AFS_PTHREAD_ENV ubik_thread_create(&urecovery_Interact_tattr, &urecovery_InteractThread, (void *)urecovery_Interact); return 0; /* is this correct? - klm */ #else code = LWP_CreateProcess(urecovery_Interact, 16384 /*8192 */ , LWP_MAX_PRIORITY - 1, (void *)0, "recovery", &junk); return code; #endif } /*! * \see ubik_ServerInitCommon() */ int ubik_ServerInitByInfo(afs_uint32 myHost, short myPort, struct afsconf_cell *info, char clones[], const char *pathName, struct ubik_dbase **dbase) { afs_int32 code; code = ubik_ServerInitCommon(myHost, myPort, info, clones, 0, pathName, dbase); return code; } /*! * \see ubik_ServerInitCommon() */ int ubik_ServerInit(afs_uint32 myHost, short myPort, afs_uint32 serverList[], const char *pathName, struct ubik_dbase **dbase) { afs_int32 code; code = ubik_ServerInitCommon(myHost, myPort, (struct afsconf_cell *)0, 0, serverList, pathName, dbase); return code; } /*! * \brief This routine begins a read or write transaction on the transaction * identified by transPtr, in the dbase named by dbase. * * An open mode of ubik_READTRANS identifies this as a read transaction, * while a mode of ubik_WRITETRANS identifies this as a write transaction. * transPtr is set to the returned transaction control block. * The readAny flag is set to 0 or 1 or 2 by the wrapper functions * ubik_BeginTrans() or ubik_BeginTransReadAny() or * ubik_BeginTransReadAnyWrite() below. * * \note We can only begin transaction when we have an up-to-date database. */ static int BeginTrans(struct ubik_dbase *dbase, afs_int32 transMode, struct ubik_trans **transPtr, int readAny) { struct ubik_trans *jt; struct ubik_trans *tt; afs_int32 code; if (readAny > 1 && ubik_SyncWriterCacheProc == NULL) { /* it's not safe to use ubik_BeginTransReadAnyWrite without a * cache-syncing function; fall back to ubik_BeginTransReadAny, * which is safe but slower */ ViceLog(0, ("ubik_BeginTransReadAnyWrite called, but " "ubik_SyncWriterCacheProc not set; pretending " "ubik_BeginTransReadAny was called instead\n")); readAny = 1; } if ((transMode != UBIK_READTRANS) && readAny) return UBADTYPE; DBHOLD(dbase); if (urecovery_AllBetter(dbase, readAny) == 0) { DBRELE(dbase); return UNOQUORUM; } /* otherwise we have a quorum, use it */ /* make sure that at most one write transaction occurs at any one time. This * has nothing to do with transaction locking; that's enforced by the lock package. However, * we can't even handle two non-conflicting writes, since our log and recovery modules * don't know how to restore one without possibly picking up some data from the other. */ if (transMode == UBIK_WRITETRANS) { /* if we're writing already, wait */ while (dbase->flags & DBWRITING) { #ifdef AFS_PTHREAD_ENV opr_cv_wait(&dbase->flags_cond, &dbase->versionLock); #else DBRELE(dbase); LWP_WaitProcess(&dbase->flags); DBHOLD(dbase); #endif } if (!ubeacon_AmSyncSite()) { DBRELE(dbase); return UNOTSYNC; } if (!ubeacon_SyncSiteAdvertised()) { /* i am the sync-site but the remotes are not aware yet */ DBRELE(dbase); return UNOQUORUM; } } /* create the transaction */ code = udisk_begin(dbase, transMode, &jt); /* can't take address of register var */ tt = jt; /* move to a register */ if (code || tt == NULL) { DBRELE(dbase); return code; } UBIK_VERSION_LOCK; if (readAny) { tt->flags |= TRREADANY; if (readAny > 1) { tt->flags |= TRREADWRITE; } } /* label trans and dbase with new tid */ tt->tid.epoch = version_globals.ubik_epochTime; /* bump by two, since tidCounter+1 means trans id'd by tidCounter has finished */ tt->tid.counter = (dbase->tidCounter += 2); if (transMode == UBIK_WRITETRANS) { /* for a write trans, we have to keep track of the write tid counter too */ dbase->writeTidCounter = tt->tid.counter; } UBIK_VERSION_UNLOCK; if (transMode == UBIK_WRITETRANS) { /* next try to start transaction on appropriate number of machines */ code = ContactQuorum_NoArguments(DISK_Begin, tt, CCheckSyncAdvertised, "DISK_Begin"); if (code) { /* we must abort the operation */ udisk_abort(tt); /* force aborts to the others */ ContactQuorum_NoArguments(DISK_Abort, tt, 0, "DISK_Abort"); udisk_end(tt); DBRELE(dbase); return code; } } *transPtr = tt; DBRELE(dbase); return 0; } /*! * \see BeginTrans() */ int ubik_BeginTrans(struct ubik_dbase *dbase, afs_int32 transMode, struct ubik_trans **transPtr) { return BeginTrans(dbase, transMode, transPtr, 0); } /*! * \see BeginTrans() */ int ubik_BeginTransReadAny(struct ubik_dbase *dbase, afs_int32 transMode, struct ubik_trans **transPtr) { return BeginTrans(dbase, transMode, transPtr, 1); } /*! * \see BeginTrans() */ int ubik_BeginTransReadAnyWrite(struct ubik_dbase *dbase, afs_int32 transMode, struct ubik_trans **transPtr) { return BeginTrans(dbase, transMode, transPtr, 2); } /*! * \brief This routine ends a read or write transaction by aborting it. */ int ubik_AbortTrans(struct ubik_trans *transPtr) { afs_int32 code; afs_int32 code2; struct ubik_dbase *dbase; dbase = transPtr->dbase; if (transPtr->flags & TRCACHELOCKED) { ReleaseReadLock(&dbase->cache_lock); transPtr->flags &= ~TRCACHELOCKED; } ObtainWriteLock(&dbase->cache_lock); DBHOLD(dbase); memset(&dbase->cachedVersion, 0, sizeof(struct ubik_version)); ReleaseWriteLock(&dbase->cache_lock); /* see if we're still up-to-date */ if (!urecovery_AllBetter(dbase, transPtr->flags & TRREADANY)) { udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); return UNOQUORUM; } if (transPtr->type == UBIK_READTRANS) { code = udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); return code; } /* below here, we know we're doing a write transaction */ if (!ubeacon_AmSyncSite()) { udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); return UNOTSYNC; } /* now it is safe to try remote abort */ code = ContactQuorum_NoArguments(DISK_Abort, transPtr, 0, "DISK_Abort"); code2 = udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); return (code ? code : code2); } static void WritebackApplicationCache(struct ubik_dbase *dbase) { int code = 0; if (ubik_SyncWriterCacheProc) { code = ubik_SyncWriterCacheProc(); } if (code) { /* we failed to sync the local cache, so just invalidate the cache; * we'll try to read the cache in again on the next read */ memset(&dbase->cachedVersion, 0, sizeof(dbase->cachedVersion)); } else { memcpy(&dbase->cachedVersion, &dbase->version, sizeof(dbase->cachedVersion)); } } /*! * \brief This routine ends a read or write transaction on the open transaction identified by transPtr. * \return an error code. */ int ubik_EndTrans(struct ubik_trans *transPtr) { afs_int32 code; struct timeval tv; afs_int32 realStart; struct ubik_server *ts; afs_int32 now; int cachelocked = 0; struct ubik_dbase *dbase; if (transPtr->type == UBIK_WRITETRANS) { code = ubik_Flush(transPtr); if (code) { ubik_AbortTrans(transPtr); return (code); } } dbase = transPtr->dbase; if (transPtr->flags & TRCACHELOCKED) { ReleaseReadLock(&dbase->cache_lock); transPtr->flags &= ~TRCACHELOCKED; } if (transPtr->type != UBIK_READTRANS) { /* must hold cache_lock before DBHOLD'ing */ ObtainWriteLock(&dbase->cache_lock); cachelocked = 1; } DBHOLD(dbase); /* give up if no longer current */ if (!urecovery_AllBetter(dbase, transPtr->flags & TRREADANY)) { udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); code = UNOQUORUM; goto error; } if (transPtr->type == UBIK_READTRANS) { /* reads are easy */ code = udisk_commit(transPtr); if (code == 0) goto success; /* update cachedVersion correctly */ udisk_end(transPtr); DBRELE(dbase); goto error; } if (!ubeacon_AmSyncSite()) { /* no longer sync site */ udisk_abort(transPtr); udisk_end(transPtr); DBRELE(dbase); code = UNOTSYNC; goto error; } /* now it is safe to do commit */ code = udisk_commit(transPtr); if (code == 0) { /* db data has been committed locally; update the local cache so * readers can get at it */ WritebackApplicationCache(dbase); ReleaseWriteLock(&dbase->cache_lock); code = ContactQuorum_NoArguments(DISK_Commit, transPtr, CStampVersion, "DISK_Commit"); } else { memset(&dbase->cachedVersion, 0, sizeof(struct ubik_version)); ReleaseWriteLock(&dbase->cache_lock); } cachelocked = 0; if (code) { /* failed to commit, so must return failure. Try to clear locks first, just for fun * Note that we don't know if this transaction will eventually commit at this point. * If it made it to a site that will be present in the next quorum, we win, otherwise * we lose. If we contact a majority of sites, then we won't be here: contacting * a majority guarantees commit, since it guarantees that one dude will be a * member of the next quorum. */ ContactQuorum_NoArguments(DISK_ReleaseLocks, transPtr, 0, "DISK_ReleaseLocks"); udisk_end(transPtr); DBRELE(dbase); goto error; } /* before we can start sending unlock messages, we must wait until all servers * that are possibly still functioning on the other side of a network partition * have timed out. Check the server structures, compute how long to wait, then * start the unlocks */ realStart = FT_ApproxTime(); while (1) { /* wait for all servers to time out */ code = 0; now = FT_ApproxTime(); /* check if we're still sync site, the guy should either come up * to us, or timeout. Put safety check in anyway */ if (now - realStart > 10 * BIGTIME) { ubik_stats.escapes++; ViceLog(0, ("ubik escaping from commit wait\n")); break; } for (ts = ubik_servers; ts; ts = ts->next) { UBIK_BEACON_LOCK; if (!ts->beaconSinceDown && now <= ts->lastBeaconSent + BIGTIME) { UBIK_BEACON_UNLOCK; /* this guy could have some damaged data, wait for him */ code = 1; tv.tv_sec = 1; /* try again after a while (ha ha) */ tv.tv_usec = 0; #ifdef AFS_PTHREAD_ENV /* we could release the dbase outside of the loop, but we do * it here, in the loop, to avoid an unnecessary RELE/HOLD * if all sites are up */ DBRELE(dbase); select(0, 0, 0, 0, &tv); DBHOLD(dbase); #else IOMGR_Select(0, 0, 0, 0, &tv); /* poll, should we wait on something? */ #endif break; } UBIK_BEACON_UNLOCK; } if (code == 0) break; /* no down ones still pseudo-active */ } /* finally, unlock all the dudes. We can return success independent of the number of servers * that really unlock the dbase; the others will do it if/when they elect a new sync site. * The transaction is committed anyway, since we succeeded in contacting a quorum * at the start (when invoking the DiskCommit function). */ ContactQuorum_NoArguments(DISK_ReleaseLocks, transPtr, 0, "DISK_ReleaseLocks"); success: udisk_end(transPtr); /* don't update cachedVersion here; it should have been updated way back * in ubik_CheckCache, and earlier in this function for writes */ DBRELE(dbase); if (cachelocked) { ReleaseWriteLock(&dbase->cache_lock); } return 0; error: if (!cachelocked) { ObtainWriteLock(&dbase->cache_lock); } memset(&dbase->cachedVersion, 0, sizeof(struct ubik_version)); ReleaseWriteLock(&dbase->cache_lock); return code; } /*! * \brief This routine reads length bytes into buffer from the current position in the database. * * The file pointer is updated appropriately (by adding the number of bytes actually transferred), and the length actually transferred is stored in the long integer pointed to by length. A short read returns zero for an error code. * * \note *length is an INOUT parameter: at the start it represents the size of the buffer, and when done, it contains the number of bytes actually transferred. */ int ubik_Read(struct ubik_trans *transPtr, void *buffer, afs_int32 length) { afs_int32 code; /* reads are easy to do: handle locally */ DBHOLD(transPtr->dbase); if (!urecovery_AllBetter(transPtr->dbase, transPtr->flags & TRREADANY)) { DBRELE(transPtr->dbase); return UNOQUORUM; } code = udisk_read(transPtr, transPtr->seekFile, buffer, transPtr->seekPos, length); if (code == 0) { transPtr->seekPos += length; } DBRELE(transPtr->dbase); return code; } /*! * \brief This routine will flush the io data in the iovec structures. * * It first flushes to the local disk and then uses ContactQuorum to write it * to the other servers. */ int ubik_Flush(struct ubik_trans *transPtr) { afs_int32 code, error = 0; if (transPtr->type != UBIK_WRITETRANS) return UBADTYPE; DBHOLD(transPtr->dbase); if (!transPtr->iovec_info.iovec_wrt_len || !transPtr->iovec_info.iovec_wrt_val) { DBRELE(transPtr->dbase); return 0; } if (!urecovery_AllBetter(transPtr->dbase, transPtr->flags & TRREADANY)) ERROR_EXIT(UNOQUORUM); if (!ubeacon_AmSyncSite()) /* only sync site can write */ ERROR_EXIT(UNOTSYNC); /* Update the rest of the servers in the quorum */ code = ContactQuorum_DISK_WriteV(transPtr, 0, &transPtr->iovec_info, &transPtr->iovec_data); if (code) { udisk_abort(transPtr); /* force aborts to the others */ ContactQuorum_NoArguments(DISK_Abort, transPtr, 0, "DISK_Abort"); transPtr->iovec_info.iovec_wrt_len = 0; transPtr->iovec_data.iovec_buf_len = 0; ERROR_EXIT(code); } /* Wrote the buffers out, so start at scratch again */ transPtr->iovec_info.iovec_wrt_len = 0; transPtr->iovec_data.iovec_buf_len = 0; error_exit: DBRELE(transPtr->dbase); return error; } int ubik_Write(struct ubik_trans *transPtr, void *vbuffer, afs_int32 length) { struct ubik_iovec *iovec; afs_int32 code, error = 0; afs_int32 pos, len, size; char * buffer = (char *)vbuffer; if (transPtr->type != UBIK_WRITETRANS) return UBADTYPE; if (!length) return 0; if (length > IOVEC_MAXBUF) { for (pos = 0, len = length; len > 0; len -= size, pos += size) { size = ((len < IOVEC_MAXBUF) ? len : IOVEC_MAXBUF); code = ubik_Write(transPtr, buffer+pos, size); if (code) return (code); } return 0; } DBHOLD(transPtr->dbase); if (!transPtr->iovec_info.iovec_wrt_val) { transPtr->iovec_info.iovec_wrt_len = 0; transPtr->iovec_info.iovec_wrt_val = malloc(IOVEC_MAXWRT * sizeof(struct ubik_iovec)); transPtr->iovec_data.iovec_buf_len = 0; transPtr->iovec_data.iovec_buf_val = malloc(IOVEC_MAXBUF); if (!transPtr->iovec_info.iovec_wrt_val || !transPtr->iovec_data.iovec_buf_val) { if (transPtr->iovec_info.iovec_wrt_val) free(transPtr->iovec_info.iovec_wrt_val); transPtr->iovec_info.iovec_wrt_val = 0; if (transPtr->iovec_data.iovec_buf_val) free(transPtr->iovec_data.iovec_buf_val); transPtr->iovec_data.iovec_buf_val = 0; DBRELE(transPtr->dbase); return UNOMEM; } } /* If this write won't fit in the structure, then flush it out and start anew */ if ((transPtr->iovec_info.iovec_wrt_len >= IOVEC_MAXWRT) || ((length + transPtr->iovec_data.iovec_buf_len) > IOVEC_MAXBUF)) { /* Can't hold the DB lock over ubik_Flush */ DBRELE(transPtr->dbase); code = ubik_Flush(transPtr); if (code) return (code); DBHOLD(transPtr->dbase); } if (!urecovery_AllBetter(transPtr->dbase, transPtr->flags & TRREADANY)) ERROR_EXIT(UNOQUORUM); if (!ubeacon_AmSyncSite()) /* only sync site can write */ ERROR_EXIT(UNOTSYNC); /* Write to the local disk */ code = udisk_write(transPtr, transPtr->seekFile, buffer, transPtr->seekPos, length); if (code) { udisk_abort(transPtr); transPtr->iovec_info.iovec_wrt_len = 0; transPtr->iovec_data.iovec_buf_len = 0; DBRELE(transPtr->dbase); return (code); } /* Collect writes for the other ubik servers (to be done in bulk) */ iovec = (struct ubik_iovec *)transPtr->iovec_info.iovec_wrt_val; iovec[transPtr->iovec_info.iovec_wrt_len].file = transPtr->seekFile; iovec[transPtr->iovec_info.iovec_wrt_len].position = transPtr->seekPos; iovec[transPtr->iovec_info.iovec_wrt_len].length = length; memcpy(&transPtr->iovec_data. iovec_buf_val[transPtr->iovec_data.iovec_buf_len], buffer, length); transPtr->iovec_info.iovec_wrt_len++; transPtr->iovec_data.iovec_buf_len += length; transPtr->seekPos += length; error_exit: DBRELE(transPtr->dbase); return error; } /*! * \brief This sets the file pointer associated with the current transaction * to the appropriate file and byte position. * * Unlike Unix files, a transaction is labelled by both a file number \p fileid * and a byte position relative to the specified file \p position. */ int ubik_Seek(struct ubik_trans *transPtr, afs_int32 fileid, afs_int32 position) { afs_int32 code; DBHOLD(transPtr->dbase); if (!urecovery_AllBetter(transPtr->dbase, transPtr->flags & TRREADANY)) { code = UNOQUORUM; } else { transPtr->seekFile = fileid; transPtr->seekPos = position; code = 0; } DBRELE(transPtr->dbase); return code; } /*! * \brief This call returns the file pointer associated with the specified * transaction in \p fileid and \p position. */ int ubik_Tell(struct ubik_trans *transPtr, afs_int32 * fileid, afs_int32 * position) { DBHOLD(transPtr->dbase); *fileid = transPtr->seekFile; *position = transPtr->seekPos; DBRELE(transPtr->dbase); return 0; } /*! * \brief This sets the file size for the currently-selected file to \p length * bytes, if length is less than the file's current size. */ int ubik_Truncate(struct ubik_trans *transPtr, afs_int32 length) { afs_int32 code, error = 0; /* Will also catch if not UBIK_WRITETRANS */ code = ubik_Flush(transPtr); if (code) return (code); DBHOLD(transPtr->dbase); /* first, check that quorum is still good, and that dbase is up-to-date */ if (!urecovery_AllBetter(transPtr->dbase, transPtr->flags & TRREADANY)) ERROR_EXIT(UNOQUORUM); if (!ubeacon_AmSyncSite()) ERROR_EXIT(UNOTSYNC); /* now do the operation locally, and propagate it out */ code = udisk_truncate(transPtr, transPtr->seekFile, length); if (!code) { code = ContactQuorum_DISK_Truncate(transPtr, 0, transPtr->seekFile, length); } if (code) { /* we must abort the operation */ udisk_abort(transPtr); /* force aborts to the others */ ContactQuorum_NoArguments(DISK_Abort, transPtr, 0, "DISK_Abort"); ERROR_EXIT(code); } error_exit: DBRELE(transPtr->dbase); return error; } /*! * \brief set a lock; all locks are released on transaction end (commit/abort) */ int ubik_SetLock(struct ubik_trans *atrans, afs_int32 apos, afs_int32 alen, int atype) { afs_int32 code = 0, error = 0; if (atype == LOCKWRITE) { if (atrans->type == UBIK_READTRANS) return UBADTYPE; code = ubik_Flush(atrans); if (code) return (code); } DBHOLD(atrans->dbase); if (atype == LOCKREAD) { code = ulock_getLock(atrans, atype, 1); if (code) ERROR_EXIT(code); } else { /* first, check that quorum is still good, and that dbase is up-to-date */ if (!urecovery_AllBetter(atrans->dbase, atrans->flags & TRREADANY)) ERROR_EXIT(UNOQUORUM); if (!ubeacon_AmSyncSite()) ERROR_EXIT(UNOTSYNC); /* now do the operation locally, and propagate it out */ code = ulock_getLock(atrans, atype, 1); if (code == 0) { code = ContactQuorum_DISK_Lock(atrans, 0, 0, 1 /*unused */ , 1 /*unused */ , LOCKWRITE); } if (code) { /* we must abort the operation */ udisk_abort(atrans); /* force aborts to the others */ ContactQuorum_NoArguments(DISK_Abort, atrans, 0, "DISK_Abort"); ERROR_EXIT(code); } } error_exit: DBRELE(atrans->dbase); return error; } /*! * \brief utility to wait for a version # to change */ int ubik_WaitVersion(struct ubik_dbase *adatabase, struct ubik_version *aversion) { DBHOLD(adatabase); while (1) { /* wait until version # changes, and then return */ if (vcmp(*aversion, adatabase->version) != 0) { DBRELE(adatabase); return 0; } #ifdef AFS_PTHREAD_ENV opr_cv_wait(&adatabase->version_cond, &adatabase->versionLock); #else DBRELE(adatabase); LWP_WaitProcess(&adatabase->version); /* same vers, just wait */ DBHOLD(adatabase); #endif } } /*! * \brief utility to get the version of the dbase a transaction is dealing with */ int ubik_GetVersion(struct ubik_trans *atrans, struct ubik_version *avers) { DBHOLD(atrans->dbase); *avers = atrans->dbase->version; DBRELE(atrans->dbase); return 0; } /*! * \brief Facility to simplify database caching. * \return zero if last trans was done on the local server and was successful. * \return -1 means bad (NULL) argument. * * If return value is non-zero and the caller is a server caching part of the * Ubik database, it should invalidate that cache. */ static int ubik_CacheUpdate(struct ubik_trans *atrans) { if (!(atrans && atrans->dbase)) return -1; return vcmp(atrans->dbase->cachedVersion, atrans->dbase->version) != 0; } /** * check and possibly update cache of ubik db. * * If the version of the cached db data is out of date, this calls (*check) to * update the cache. If (*check) returns success, we update the version of the * cached db data. * * Checking the version of the cached db data is done under a read lock; * updating the cache (and thus calling (*check)) is done under a write lock * so is guaranteed not to interfere with another thread's (*check). On * successful return, a read lock on the cached db data is obtained, which * will be released by ubik_EndTrans or ubik_AbortTrans. * * @param[in] atrans ubik transaction * @param[in] check function to call to check/update cache * @param[in] rock rock to pass to *check * * @return operation status * @retval 0 success * @retval nonzero error; cachedVersion not updated * * @post On success, application cache is read-locked, and cache data is * up-to-date */ int ubik_CheckCache(struct ubik_trans *atrans, ubik_updatecache_func cbf, void *rock) { int ret = 0; if (!(atrans && atrans->dbase)) return -1; ObtainReadLock(&atrans->dbase->cache_lock); while (ubik_CacheUpdate(atrans) != 0) { ReleaseReadLock(&atrans->dbase->cache_lock); ObtainSharedLock(&atrans->dbase->cache_lock); if (ubik_CacheUpdate(atrans) != 0) { BoostSharedLock(&atrans->dbase->cache_lock); ret = (*cbf) (atrans, rock); if (ret == 0) { memcpy(&atrans->dbase->cachedVersion, &atrans->dbase->version, sizeof(atrans->dbase->cachedVersion)); } } /* It would be nice if we could convert from a shared lock to a read * lock... instead, just release the shared and acquire the read */ ReleaseSharedLock(&atrans->dbase->cache_lock); if (ret) { /* if we have an error, don't retry, and don't hold any locks */ return ret; } ObtainReadLock(&atrans->dbase->cache_lock); } atrans->flags |= TRCACHELOCKED; return 0; } /*! * "Who said anything about panicking?" snapped Arthur. * "This is still just the culture shock. You wait till I've settled down * into the situation and found my bearings. \em Then I'll start panicking!" * --Authur Dent * * \returns There is no return from panic. */ void panic(char *format, ...) { va_list ap; va_start(ap, format); ViceLog(0, ("Ubik PANIC:\n")); vViceLog(0, (format, ap)); va_end(ap); abort(); AFS_UNREACHED(ViceLog(0, ("BACK FROM ABORT\n"))); AFS_UNREACHED(exit(1)); } /*! * This function takes an IP addresses as its parameter. It returns the * the primary IP address that is on the host passed in, or 0 if not found. */ afs_uint32 ubikGetPrimaryInterfaceAddr(afs_uint32 addr) { struct ubik_server *ts; int j; UBIK_ADDR_LOCK; for (ts = ubik_servers; ts; ts = ts->next) for (j = 0; j < UBIK_MAX_INTERFACE_ADDR; j++) if (ts->addr[j] == addr) { UBIK_ADDR_UNLOCK; return ts->addr[0]; /* net byte order */ } UBIK_ADDR_UNLOCK; return 0; /* if not in server database, return error */ } int ubik_CheckAuth(struct rx_call *acall) { if (checkSecurityProc) return (*checkSecurityProc) (securityRock, acall); else if (ubik_CheckRXSecurityProc) { return (*ubik_CheckRXSecurityProc) (ubik_CheckRXSecurityRock, acall); } else return 0; } void ubik_SetServerSecurityProcs(void (*buildproc) (void *, struct rx_securityClass ***, afs_int32 *), int (*checkproc) (void *, struct rx_call *), void *rock) { buildSecClassesProc = buildproc; checkSecurityProc = checkproc; securityRock = rock; }