Import Upstream version 1.8.5

[hcoop/debian/openafs.git] / src / ubik / recovery.c

/*
 * 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 <afsconfig.h>
#include <afs/param.h>

#include <roken.h>

#include <afs/opr.h>

#ifdef AFS_PTHREAD_ENV
# include <opr/lock.h>
#else
# include <opr/lockstub.h>
#endif

#include <rx/rx.h>
#include <afs/afsutil.h>
#include <afs/cellconfig.h>


#define UBIK_INTERNALS
#include "ubik.h"
#include "ubik_int.h"

/*! \file
 * This module is responsible for determining when the system has
 * recovered to the point that it can handle new transactions.  It
 * replays logs, polls to determine the current dbase after a crash,
 * and distributes the new database to the others.
 *
 * The sync site associates a version number with each database.  It
 * broadcasts the version associated with its current dbase in every
 * one of its beacon messages.  When the sync site send a dbase to a
 * server, it also sends the db's version.  A non-sync site server can
 * tell if it has the right dbase version by simply comparing the
 * version from the beacon message \p uvote_dbVersion with the version
 * associated with the database \p ubik_dbase->version.  The sync site
 * itself simply has one counter to keep track of all of this (again
 * \p ubik_dbase->version).
 *
 * sync site: routine called when the sync site loses its quorum; this
 * procedure is called "up" from the beacon package.  It resyncs the
 * dbase and nudges the recovery daemon to try to propagate out the
 * changes.  It also resets the recovery daemon's state, since
 * recovery must potentially find a new dbase to propagate out.  This
 * routine should not do anything with variables used by non-sync site
 * servers.
 */

/*!
 * if this flag is set, then ubik will use only the primary address
 * (the address specified in the CellServDB) to contact other
 * ubik servers. Ubik recovery will not try opening connections
 * to the alternate interface addresses.
 */
int ubikPrimaryAddrOnly;

int
urecovery_ResetState(void)
{
    urecovery_state = 0;
    return 0;
}

/*!
 * \brief sync site
 *
 * routine called when a non-sync site server goes down; restarts recovery
 * process to send missing server the new db when it comes back up for
 * non-sync site servers.
 *
 * \note This routine should not do anything with variables used by non-sync site servers.
 */
int
urecovery_LostServer(struct ubik_server *ts)
{
    ubeacon_ReinitServer(ts);
    return 0;
}

/*!
 * return true iff we have a current database (called by both sync
 * sites and non-sync sites) How do we determine this?  If we're the
 * sync site, we wait until recovery has finished fetching and
 * re-labelling its dbase (it may still be trying to propagate it out
 * to everyone else; that's THEIR problem).  If we're not the sync
 * site, then we must have a dbase labelled with the right version,
 * and we must have a currently-good sync site.
 */
int
urecovery_AllBetter(struct ubik_dbase *adbase, int areadAny)
{
    afs_int32 rcode;

    ViceLog(25, ("allbetter checking\n"));
    rcode = 0;


    if (areadAny) {
        if (ubik_dbase->version.epoch > 1)
            rcode = 1;          /* Happy with any good version of database */
    }

    /* Check if we're sync site and we've got the right data */
    else if (ubeacon_AmSyncSite() && (urecovery_state & UBIK_RECHAVEDB)) {
        rcode = 1;
    }

    /* next, check if we're aux site, and we've ever been sent the
     * right data (note that if a dbase update fails, we won't think
     * that the sync site is still the sync site, 'cause it won't talk
     * to us until a timeout period has gone by.  When we recover, we
     * leave this clear until we get a new dbase */
    else if (uvote_HaveSyncAndVersion(ubik_dbase->version)) {
        rcode = 1;
    }

    ViceLog(25, ("allbetter: returning %d\n", rcode));
    return rcode;
}

/*!
 * \brief abort all transactions on this database
 */
int
urecovery_AbortAll(struct ubik_dbase *adbase)
{
    struct ubik_trans *tt;
    for (tt = adbase->activeTrans; tt; tt = tt->next) {
        udisk_abort(tt);
    }
    return 0;
}

/*!
 * \brief this routine aborts the current remote transaction, if any, if the tid is wrong
 */
int
urecovery_CheckTid(struct ubik_tid *atid, int abortalways)
{
    if (ubik_currentTrans) {
        /* there is remote write trans, see if we match, see if this
         * is a new transaction */
        if (atid->epoch != ubik_currentTrans->tid.epoch
            || atid->counter > ubik_currentTrans->tid.counter || abortalways) {
            /* don't match, abort it */
            /* If the thread is not waiting for lock - ok to end it */
            if (ubik_currentTrans->locktype != LOCKWAIT) {
                udisk_end(ubik_currentTrans);
            }
            ubik_currentTrans = (struct ubik_trans *)0;
        }
    }
    return 0;
}

/*!
 * \brief replay logs
 *
 * log format is defined here, and implicitly in disk.c
 *
 * 4 byte opcode, followed by parameters, each 4 bytes long.  All integers
 * are in logged in network standard byte order, in case we want to move logs
 * from machine-to-machine someday.
 *
 * Begin transaction: opcode \n
 * Commit transaction: opcode, version (8 bytes) \n
 * Truncate file: opcode, file number, length \n
 * Abort transaction: opcode \n
 * Write data: opcode, file, position, length, <length> data bytes \n
 *
 * A very simple routine, it just replays the log.  Note that this is a new-value only log, which
 * implies that no uncommitted data is written to the dbase: one writes data to the log, including
 * the commit record, then we allow data to be written through to the dbase.  In our particular
 * implementation, once a transaction is done, we write out the pages to the database, so that
 * our buffer package doesn't have to know about stable and uncommitted data in the memory buffers:
 * any changed data while there is an uncommitted write transaction can be zapped during an
 * abort and the remaining dbase on the disk is exactly the right dbase, without having to read
 * the log.
 */
static int
ReplayLog(struct ubik_dbase *adbase)
{
    afs_int32 opcode;
    afs_int32 code, tpos;
    int logIsGood;
    afs_int32 len, thisSize, tfile, filePos;
    afs_int32 buffer[4];
    afs_int32 syncFile = -1;
    afs_int32 data[1024];

    /* read the lock twice, once to see whether we have a transaction to deal
     * with that committed, (theoretically, we should support more than one
     * trans in the log at once, but not yet), and once replaying the
     * transactions.  */
    tpos = 0;
    logIsGood = 0;
    /* for now, assume that all ops in log pertain to one transaction; see if there's a commit */
    while (1) {
        code =
            (*adbase->read) (adbase, LOGFILE, (char *)&opcode, tpos,
                             sizeof(afs_int32));
        if (code != sizeof(afs_int32))
            break;
        opcode = ntohl(opcode);
        if (opcode == LOGNEW) {
            /* handle begin trans */
            tpos += sizeof(afs_int32);
        } else if (opcode == LOGABORT)
            break;
        else if (opcode == LOGEND) {
            logIsGood = 1;
            break;
        } else if (opcode == LOGTRUNCATE) {
            tpos += 4;
            code =
                (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                 2 * sizeof(afs_int32));
            if (code != 2 * sizeof(afs_int32))
                break;          /* premature eof or io error */
            tpos += 2 * sizeof(afs_int32);
        } else if (opcode == LOGDATA) {
            tpos += 4;
            code =
                (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                 3 * sizeof(afs_int32));
            if (code != 3 * sizeof(afs_int32))
                break;
            /* otherwise, skip over the data bytes, too */
            tpos += ntohl(buffer[2]) + 3 * sizeof(afs_int32);
        } else {
            ViceLog(0, ("corrupt log opcode (%d) at position %d\n", opcode,
                       tpos));
            break;              /* corrupt log! */
        }
    }
    if (logIsGood) {
        /* actually do the replay; log should go all the way through the commit record, since
         * we just read it above. */
        tpos = 0;
        logIsGood = 0;
        syncFile = -1;
        while (1) {
            code =
                (*adbase->read) (adbase, LOGFILE, (char *)&opcode, tpos,
                                 sizeof(afs_int32));
            if (code != sizeof(afs_int32))
                break;
            opcode = ntohl(opcode);
            if (opcode == LOGNEW) {
                /* handle begin trans */
                tpos += sizeof(afs_int32);
            } else if (opcode == LOGABORT)
                panic("log abort\n");
            else if (opcode == LOGEND) {
                struct ubik_version version;
                tpos += 4;
                code =
                    (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                     2 * sizeof(afs_int32));
                if (code != 2 * sizeof(afs_int32))
                    return UBADLOG;
                version.epoch = ntohl(buffer[0]);
                version.counter = ntohl(buffer[1]);
                code = (*adbase->setlabel) (adbase, 0, &version);
                if (code)
                    return code;
                ViceLog(0, ("Successfully replayed log for interrupted "
                           "transaction; db version is now %ld.%ld\n",
                           (long) version.epoch, (long) version.counter));
                logIsGood = 1;
                break;          /* all done now */
            } else if (opcode == LOGTRUNCATE) {
                tpos += 4;
                code =
                    (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                     2 * sizeof(afs_int32));
                if (code != 2 * sizeof(afs_int32))
                    break;      /* premature eof or io error */
                tpos += 2 * sizeof(afs_int32);
                code =
                    (*adbase->truncate) (adbase, ntohl(buffer[0]),
                                         ntohl(buffer[1]));
                if (code)
                    return code;
            } else if (opcode == LOGDATA) {
                tpos += 4;
                code =
                    (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                     3 * sizeof(afs_int32));
                if (code != 3 * sizeof(afs_int32))
                    break;
                tpos += 3 * sizeof(afs_int32);
                /* otherwise, skip over the data bytes, too */
                len = ntohl(buffer[2]); /* total number of bytes to copy */
                filePos = ntohl(buffer[1]);
                tfile = ntohl(buffer[0]);
                /* try to minimize file syncs */
                if (syncFile != tfile) {
                    if (syncFile >= 0)
                        code = (*adbase->sync) (adbase, syncFile);
                    else
                        code = 0;
                    syncFile = tfile;
                    if (code)
                        return code;
                }
                while (len > 0) {
                    thisSize = (len > sizeof(data) ? sizeof(data) : len);
                    /* copy sizeof(data) buffer bytes at a time */
                    code =
                        (*adbase->read) (adbase, LOGFILE, (char *)data, tpos,
                                         thisSize);
                    if (code != thisSize)
                        return UBADLOG;
                    code =
                        (*adbase->write) (adbase, tfile, (char *)data, filePos,
                                          thisSize);
                    if (code != thisSize)
                        return UBADLOG;
                    filePos += thisSize;
                    tpos += thisSize;
                    len -= thisSize;
                }
            } else {
                ViceLog(0, ("corrupt log opcode (%d) at position %d\n",
                           opcode, tpos));
                break;          /* corrupt log! */
            }
        }
        if (logIsGood) {
            if (syncFile >= 0)
                code = (*adbase->sync) (adbase, syncFile);
            if (code)
                return code;
        } else {
            ViceLog(0, ("Log read error on pass 2\n"));
            return UBADLOG;
        }
    }

    /* now truncate the log, we're done with it */
    code = (*adbase->truncate) (adbase, LOGFILE, 0);
    return code;
}

/*! \brief
 * Called at initialization to figure out version of the dbase we really have.
 *
 * This routine is called after replaying the log; it reads the restored labels.
 */
static int
InitializeDB(struct ubik_dbase *adbase)
{
    afs_int32 code;

    code = (*adbase->getlabel) (adbase, 0, &adbase->version);
    if (code) {
        /* try setting the label to a new value */
        UBIK_VERSION_LOCK;
        adbase->version.epoch = 1;      /* value for newly-initialized db */
        adbase->version.counter = 1;
        code = (*adbase->setlabel) (adbase, 0, &adbase->version);
        if (code) {
            /* failed, try to set it back */
            adbase->version.epoch = 0;
            adbase->version.counter = 0;
            (*adbase->setlabel) (adbase, 0, &adbase->version);
        }
#ifdef AFS_PTHREAD_ENV
        opr_cv_broadcast(&adbase->version_cond);
#else
        LWP_NoYieldSignal(&adbase->version);
#endif
        UBIK_VERSION_UNLOCK;
    }
    return 0;
}

/*!
 * \brief initialize the local ubik_dbase
 *
 * We replay the logs and then read the resulting file to figure out what version we've really got.
 */
int
urecovery_Initialize(struct ubik_dbase *adbase)
{
    afs_int32 code;

    DBHOLD(adbase);
    code = ReplayLog(adbase);
    if (code)
        goto done;
    code = InitializeDB(adbase);
done:
    DBRELE(adbase);
    return code;
}

/*!
 * \brief Main interaction loop for the recovery manager
 *
 * The recovery light-weight process only runs when you're the
 * synchronization site.  It performs the following tasks, if and only
 * if the prerequisite tasks have been performed successfully (it
 * keeps track of which ones have been performed in its bit map,
 * \p urecovery_state).
 *
 * First, it is responsible for probing that all servers are up.  This
 * is the only operation that must be performed even if this is not
 * yet the sync site, since otherwise this site may not notice that
 * enough other machines are running to even elect this guy to be the
 * sync site.
 *
 * After that, the recovery process does nothing until the beacon and
 * voting modules manage to get this site elected sync site.
 *
 * After becoming sync site, recovery first attempts to find the best
 * database available in the network (it must do this in order to
 * ensure finding the latest committed data).  After finding the right
 * database, it must fetch this dbase to the sync site.
 *
 * After fetching the dbase, it relabels it with a new version number,
 * to ensure that everyone recognizes this dbase as the most recent
 * dbase.
 *
 * One the dbase has been relabelled, this machine can start handling
 * requests.  However, the recovery module still has one more task:
 * propagating the dbase out to everyone who is up in the network.
 */
void *
urecovery_Interact(void *dummy)
{
    afs_int32 code;
    struct ubik_server *bestServer = NULL;
    struct ubik_server *ts;
    int dbok, doingRPC, now;
    afs_int32 lastProbeTime;
    /* if we're the sync site, the best db version we've found yet */
    static struct ubik_version bestDBVersion;
    struct ubik_version tversion;
    struct timeval tv;
    int length, tlen, offset, file, nbytes;
    struct rx_call *rxcall;
    char tbuffer[1024];
    struct ubik_stat ubikstat;
    struct in_addr inAddr;
    char hoststr[16];
    char pbuffer[1028];
    int fd = -1;
    afs_int32 pass;

    afs_pthread_setname_self("recovery");

    /* otherwise, begin interaction */
    urecovery_state = 0;
    lastProbeTime = 0;
    while (1) {
        /* Run through this loop every 4 seconds */
        tv.tv_sec = 4;
        tv.tv_usec = 0;
#ifdef AFS_PTHREAD_ENV
        select(0, 0, 0, 0, &tv);
#else
        IOMGR_Select(0, 0, 0, 0, &tv);
#endif

        ViceLog(5, ("recovery running in state %x\n", urecovery_state));

        /* Every 30 seconds, check all the down servers and mark them
         * as up if they respond. When a server comes up or found to
         * not be current, then re-find the the best database and
         * propogate it.
         */
        if ((now = FT_ApproxTime()) > 30 + lastProbeTime) {

            for (ts = ubik_servers, doingRPC = 0; ts; ts = ts->next) {
                UBIK_BEACON_LOCK;
                if (!ts->up) {
                    UBIK_BEACON_UNLOCK;
                    doingRPC = 1;
                    code = DoProbe(ts);
                    if (code == 0) {
                        UBIK_BEACON_LOCK;
                        ts->up = 1;
                        UBIK_BEACON_UNLOCK;
                        DBHOLD(ubik_dbase);
                        urecovery_state &= ~UBIK_RECFOUNDDB;
                        DBRELE(ubik_dbase);
                    }
                } else {
                    UBIK_BEACON_UNLOCK;
                    DBHOLD(ubik_dbase);
                    if (!ts->currentDB)
                        urecovery_state &= ~UBIK_RECFOUNDDB;
                    DBRELE(ubik_dbase);
                }
            }

            if (doingRPC)
                now = FT_ApproxTime();
            lastProbeTime = now;
        }

        /* Mark whether we are the sync site */
        DBHOLD(ubik_dbase);
        if (!ubeacon_AmSyncSite()) {
            urecovery_state &= ~UBIK_RECSYNCSITE;
            DBRELE(ubik_dbase);
            continue;           /* nothing to do */
        }
        urecovery_state |= UBIK_RECSYNCSITE;

        /* If a server has just come up or if we have not found the
         * most current database, then go find the most current db.
         */
        if (!(urecovery_state & UBIK_RECFOUNDDB)) {
            int okcalls = 0;
            DBRELE(ubik_dbase);
            bestServer = (struct ubik_server *)0;
            bestDBVersion.epoch = 0;
            bestDBVersion.counter = 0;
            for (ts = ubik_servers; ts; ts = ts->next) {
                UBIK_BEACON_LOCK;
                if (!ts->up) {
                    UBIK_BEACON_UNLOCK;
                    continue;   /* don't bother with these guys */
                }
                UBIK_BEACON_UNLOCK;
                if (ts->isClone)
                    continue;
                UBIK_ADDR_LOCK;
                code = DISK_GetVersion(ts->disk_rxcid, &ts->version);
                UBIK_ADDR_UNLOCK;
                if (code == 0) {
                    okcalls++;
                    /* perhaps this is the best version */
                    if (vcmp(ts->version, bestDBVersion) > 0) {
                        /* new best version */
                        bestDBVersion = ts->version;
                        bestServer = ts;
                    }
                }
            }

            DBHOLD(ubik_dbase);

            if (okcalls + 1 >= ubik_quorum) {
                /* If we've asked a majority of sites about their db version,
                 * then we can say with confidence that we've found the best db
                 * version. If we haven't contacted most sites (because
                 * GetVersion failed or because we already know the server is
                 * down), then we don't really know if we know about the best
                 * db version. So we can only proceed in here if 'okcalls'
                 * indicates we managed to contact a majority of sites. */

                /* take into consideration our version. Remember if we,
                 * the sync site, have the best version. Also note that
                 * we may need to send the best version out.
                 */
                if (vcmp(ubik_dbase->version, bestDBVersion) >= 0) {
                    bestDBVersion = ubik_dbase->version;
                    bestServer = (struct ubik_server *)0;
                    urecovery_state |= UBIK_RECHAVEDB;
                } else {
                    /* Clear the flag only when we know we have to retrieve
                     * the db. Because urecovery_AllBetter() looks at it.
                     */
                    urecovery_state &= ~UBIK_RECHAVEDB;
                }
                urecovery_state |= UBIK_RECFOUNDDB;
                urecovery_state &= ~UBIK_RECSENTDB;
            }
        }
        if (!(urecovery_state & UBIK_RECFOUNDDB)) {
            DBRELE(ubik_dbase);
            continue;           /* not ready */
        }

        /* If we, the sync site, do not have the best db version, then
         * go and get it from the server that does.
         */
        if ((urecovery_state & UBIK_RECHAVEDB) || !bestServer) {
            urecovery_state |= UBIK_RECHAVEDB;
        } else {
            /* we don't have the best version; we should fetch it. */
            urecovery_AbortAll(ubik_dbase);

            /* Rx code to do the Bulk fetch */
            file = 0;
            offset = 0;
            UBIK_ADDR_LOCK;
            rxcall = rx_NewCall(bestServer->disk_rxcid);

            ViceLog(0, ("Ubik: Synchronize database via DISK_GetFile to server %s\n",
                       afs_inet_ntoa_r(bestServer->addr[0], hoststr)));
            UBIK_ADDR_UNLOCK;

            code = StartDISK_GetFile(rxcall, file);
            if (code) {
                ViceLog(5, ("StartDiskGetFile failed=%d\n", code));
                goto FetchEndCall;
            }
            nbytes = rx_Read(rxcall, (char *)&length, sizeof(afs_int32));
            length = ntohl(length);
            if (nbytes != sizeof(afs_int32)) {
                ViceLog(5, ("Rx-read length error=%d\n", BULK_ERROR));
                code = EIO;
                goto FetchEndCall;
            }

            /* give invalid label during file transit */
            UBIK_VERSION_LOCK;
            tversion.epoch = 0;
            code = (*ubik_dbase->setlabel) (ubik_dbase, file, &tversion);
            UBIK_VERSION_UNLOCK;
            if (code) {
                ViceLog(5, ("setlabel io error=%d\n", code));
                goto FetchEndCall;
            }
            snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
                     ubik_dbase->pathName, (file<0)?"SYS":"",
                     (file<0)?-file:file);
            fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
            if (fd < 0) {
                code = errno;
                goto FetchEndCall;
            }
            code = lseek(fd, HDRSIZE, 0);
            if (code != HDRSIZE) {
                close(fd);
                goto FetchEndCall;
            }

            pass = 0;
            while (length > 0) {
                tlen = (length > sizeof(tbuffer) ? sizeof(tbuffer) : length);
#ifndef AFS_PTHREAD_ENV
                if (pass % 4 == 0)
                    IOMGR_Poll();
#endif
                nbytes = rx_Read(rxcall, tbuffer, tlen);
                if (nbytes != tlen) {
                    ViceLog(5, ("Rx-read bulk error=%d\n", BULK_ERROR));
                    code = EIO;
                    close(fd);
                    goto FetchEndCall;
                }
                nbytes = write(fd, tbuffer, tlen);
                pass++;
                if (nbytes != tlen) {
                    code = UIOERROR;
                    close(fd);
                    goto FetchEndCall;
                }
                offset += tlen;
                length -= tlen;
            }
            code = close(fd);
            if (code)
                goto FetchEndCall;
            code = EndDISK_GetFile(rxcall, &tversion);
          FetchEndCall:
            code = rx_EndCall(rxcall, code);
            if (!code) {
                /* we got a new file, set up its header */
                urecovery_state |= UBIK_RECHAVEDB;
                UBIK_VERSION_LOCK;
                memcpy(&ubik_dbase->version, &tversion,
                       sizeof(struct ubik_version));
                snprintf(tbuffer, sizeof(tbuffer), "%s.DB%s%d",
                         ubik_dbase->pathName, (file<0)?"SYS":"",
                         (file<0)?-file:file);
#ifdef AFS_NT40_ENV
                snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
                         ubik_dbase->pathName, (file<0)?"SYS":"",
                         (file<0)?-file:file);
                code = unlink(pbuffer);
                if (!code)
                    code = rename(tbuffer, pbuffer);
                snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
                         ubik_dbase->pathName, (file<0)?"SYS":"",
                         (file<0)?-file:file);
#endif
                if (!code)
                    code = rename(pbuffer, tbuffer);
                if (!code) {
                    (*ubik_dbase->open) (ubik_dbase, file);
                    /* after data is good, sync disk with correct label */
                    code =
                        (*ubik_dbase->setlabel) (ubik_dbase, 0,
                                                 &ubik_dbase->version);
                }
                UBIK_VERSION_UNLOCK;
#ifdef AFS_NT40_ENV
                snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.OLD",
                         ubik_dbase->pathName, (file<0)?"SYS":"",
                         (file<0)?-file:file);
                unlink(pbuffer);
#endif
            }
            if (code) {
                unlink(pbuffer);
                /*
                 * We will effectively invalidate the old data forever now.
                 * Unclear if we *should* but we do.
                 */
                UBIK_VERSION_LOCK;
                ubik_dbase->version.epoch = 0;
                ubik_dbase->version.counter = 0;
                UBIK_VERSION_UNLOCK;
                ViceLog(0, ("Ubik: Synchronize database failed (error = %d)\n",
                           code));
            } else {
                ViceLog(0, ("Ubik: Synchronize database completed\n"));
                urecovery_state |= UBIK_RECHAVEDB;
            }
            udisk_Invalidate(ubik_dbase, 0);    /* data has changed */
#ifdef AFS_PTHREAD_ENV
            opr_cv_broadcast(&ubik_dbase->version_cond);
#else
            LWP_NoYieldSignal(&ubik_dbase->version);
#endif
        }
        if (!(urecovery_state & UBIK_RECHAVEDB)) {
            DBRELE(ubik_dbase);
            continue;           /* not ready */
        }

        /* If the database was newly initialized, then when we establish quorum, write
         * a new label. This allows urecovery_AllBetter() to allow access for reads.
         * Setting it to 2 also allows another site to come along with a newer
         * database and overwrite this one.
         */
        if (ubik_dbase->version.epoch == 1) {
            urecovery_AbortAll(ubik_dbase);
            UBIK_VERSION_LOCK;
            ubik_dbase->version.epoch = 2;
            ubik_dbase->version.counter = 1;
            code =
                (*ubik_dbase->setlabel) (ubik_dbase, 0, &ubik_dbase->version);
            UBIK_VERSION_UNLOCK;
            udisk_Invalidate(ubik_dbase, 0);    /* data may have changed */
#ifdef AFS_PTHREAD_ENV
            opr_cv_broadcast(&ubik_dbase->version_cond);
#else
            LWP_NoYieldSignal(&ubik_dbase->version);
#endif
        }

        /* Check the other sites and send the database to them if they
         * do not have the current db.
         */
        if (!(urecovery_state & UBIK_RECSENTDB)) {
            /* now propagate out new version to everyone else */
            dbok = 1;           /* start off assuming they all worked */

            /*
             * Check if a write transaction is in progress. We can't send the
             * db when a write is in progress here because the db would be
             * obsolete as soon as it goes there. Also, ops after the begin
             * trans would reach the recepient and wouldn't find a transaction
             * pending there.  Frankly, I don't think it's possible to get past
             * the write-lock above if there is a write transaction in progress,
             * but then, it won't hurt to check, will it?
             */
            if (ubik_dbase->flags & DBWRITING) {
                struct timeval tv;
                int safety = 0;
                long cur_usec = 50000;
                while ((ubik_dbase->flags & DBWRITING) && (safety < 500)) {
                    DBRELE(ubik_dbase);
                    /* sleep for a little while */
                    tv.tv_sec = 0;
                    tv.tv_usec = cur_usec;
#ifdef AFS_PTHREAD_ENV
                    select(0, 0, 0, 0, &tv);
#else
                    IOMGR_Select(0, 0, 0, 0, &tv);
#endif
                    cur_usec += 10000;
                    safety++;
                    DBHOLD(ubik_dbase);
                }
            }

            for (ts = ubik_servers; ts; ts = ts->next) {
                UBIK_ADDR_LOCK;
                inAddr.s_addr = ts->addr[0];
                UBIK_ADDR_UNLOCK;
                UBIK_BEACON_LOCK;
                if (!ts->up) {
                    UBIK_BEACON_UNLOCK;
                    ViceLog(5, ("recovery cannot send version to %s\n",
                                afs_inet_ntoa_r(inAddr.s_addr, hoststr)));
                    dbok = 0;
                    continue;
                }
                UBIK_BEACON_UNLOCK;
                ViceLog(5, ("recovery sending version to %s\n",
                            afs_inet_ntoa_r(inAddr.s_addr, hoststr)));
                if (vcmp(ts->version, ubik_dbase->version) != 0) {
                    ViceLog(5, ("recovery stating local database\n"));

                    /* Rx code to do the Bulk Store */
                    code = (*ubik_dbase->stat) (ubik_dbase, 0, &ubikstat);
                    if (!code) {
                        length = ubikstat.size;
                        file = offset = 0;
                        UBIK_ADDR_LOCK;
                        rxcall = rx_NewCall(ts->disk_rxcid);
                        UBIK_ADDR_UNLOCK;
                        code =
                            StartDISK_SendFile(rxcall, file, length,
                                               &ubik_dbase->version);
                        if (code) {
                            ViceLog(5, ("StartDiskSendFile failed=%d\n",
                                        code));
                            goto StoreEndCall;
                        }
                        while (length > 0) {
                            tlen =
                                (length >
                                 sizeof(tbuffer) ? sizeof(tbuffer) : length);
                            nbytes =
                                (*ubik_dbase->read) (ubik_dbase, file,
                                                     tbuffer, offset, tlen);
                            if (nbytes != tlen) {
                                code = UIOERROR;
                                ViceLog(5, ("Local disk read error=%d\n", code));
                                goto StoreEndCall;
                            }
                            nbytes = rx_Write(rxcall, tbuffer, tlen);
                            if (nbytes != tlen) {
                                code = BULK_ERROR;
                                ViceLog(5, ("Rx-write bulk error=%d\n", code));
                                goto StoreEndCall;
                            }
                            offset += tlen;
                            length -= tlen;
                        }
                        code = EndDISK_SendFile(rxcall);
                      StoreEndCall:
                        code = rx_EndCall(rxcall, code);
                    }
                    if (code == 0) {
                        /* we set a new file, process its header */
                        ts->version = ubik_dbase->version;
                        ts->currentDB = 1;
                    } else
                        dbok = 0;
                } else {
                    /* mark file up to date */
                    ts->currentDB = 1;
                }
            }
            if (dbok)
                urecovery_state |= UBIK_RECSENTDB;
        }
        DBRELE(ubik_dbase);
    }
    AFS_UNREACHED(return(NULL));
}

/*!
 * \brief send a Probe to all the network address of this server
 *
 * \return 0 if success, else return 1
 */
int
DoProbe(struct ubik_server *server)
{
    struct rx_connection *conns[UBIK_MAX_INTERFACE_ADDR];
    struct rx_connection *connSuccess = 0;
    int i, nconns, success_i = -1;
    afs_uint32 addr;
    char buffer[32];
    char hoststr[16];

    UBIK_ADDR_LOCK;
    for (i = 0; (i < UBIK_MAX_INTERFACE_ADDR) && (addr = server->addr[i]);
         i++) {
        conns[i] =
            rx_NewConnection(addr, ubik_callPortal, DISK_SERVICE_ID,
                             addr_globals.ubikSecClass, addr_globals.ubikSecIndex);

        /* user requirement to use only the primary interface */
        if (ubikPrimaryAddrOnly) {
            i = 1;
            break;
        }
    }
    UBIK_ADDR_UNLOCK;
    nconns = i;
    opr_Assert(nconns);                 /* at least one interface address for this server */

    multi_Rx(conns, nconns) {
        multi_DISK_Probe();
        if (!multi_error) {     /* first success */
            success_i = multi_i;

            multi_Abort;
        }
    } multi_End_Ignore;

    if (success_i >= 0) {
        UBIK_ADDR_LOCK;
        addr = server->addr[success_i]; /* successful interface addr */

        if (server->disk_rxcid) /* destroy existing conn */
            rx_DestroyConnection(server->disk_rxcid);
        if (server->vote_rxcid)
            rx_DestroyConnection(server->vote_rxcid);

        /* make new connections */
        server->disk_rxcid = conns[success_i];
        server->vote_rxcid = rx_NewConnection(addr, ubik_callPortal,
                                              VOTE_SERVICE_ID, addr_globals.ubikSecClass,
                                              addr_globals.ubikSecIndex);

        connSuccess = conns[success_i];
        strcpy(buffer, afs_inet_ntoa_r(server->addr[0], hoststr));

        ViceLog(0, ("ubik:server %s is back up: will be contacted through %s\n",
             buffer, afs_inet_ntoa_r(addr, hoststr)));
        UBIK_ADDR_UNLOCK;
    }

    /* Destroy all connections except the one on which we succeeded */
    for (i = 0; i < nconns; i++)
        if (conns[i] != connSuccess)
            rx_DestroyConnection(conns[i]);

    if (!connSuccess)
        ViceLog(5, ("ubik:server %s still down\n",
                    afs_inet_ntoa_r(server->addr[0], hoststr)));

    if (connSuccess)
        return 0;               /* success */
    else
        return 1;               /* failure */
}