#include <roken.h>
-#include <lock.h>
-#include <rx/xdr.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"
urecovery_ResetState(void)
{
urecovery_state = 0;
-#if !defined(AFS_PTHREAD_ENV)
- /* No corresponding LWP_WaitProcess found anywhere for this -- klm */
- LWP_NoYieldSignal(&urecovery_state);
-#endif
return 0;
}
urecovery_LostServer(struct ubik_server *ts)
{
ubeacon_ReinitServer(ts);
-#if !defined(AFS_PTHREAD_ENV)
- /* No corresponding LWP_WaitProcess found anywhere for this -- klm */
- LWP_NoYieldSignal(&urecovery_state);
-#endif
return 0;
}
{
afs_int32 rcode;
- ubik_dprint_25("allbetter checking\n");
+ ViceLog(25, ("allbetter checking\n"));
rcode = 0;
* 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_GetSyncSite() && uvote_eq_dbVersion(ubik_dbase->version))) { /* && order is important */
+ else if (uvote_HaveSyncAndVersion(ubik_dbase->version)) {
rcode = 1;
}
- ubik_dprint_25("allbetter: returning %d\n", rcode);
+ ViceLog(25, ("allbetter: returning %d\n", rcode));
return rcode;
}
urecovery_AbortAll(struct ubik_dbase *adbase)
{
struct ubik_trans *tt;
+ int reads = 0, writes = 0;
+
for (tt = adbase->activeTrans; tt; tt = tt->next) {
+ if (tt->type == UBIK_WRITETRANS)
+ writes++;
+ else
+ reads++;
udisk_abort(tt);
}
+ ViceLog(0, ("urecovery_AbortAll: just aborted %d read and %d write transactions\n",
+ reads, writes));
return 0;
}
if (atid->epoch != ubik_currentTrans->tid.epoch
|| atid->counter > ubik_currentTrans->tid.counter || abortalways) {
/* don't match, abort it */
+ int endit = 0;
/* If the thread is not waiting for lock - ok to end it */
if (ubik_currentTrans->locktype != LOCKWAIT) {
+ endit = 1;
+ }
+
+ ViceLog(0, ("urecovery_CheckTid: Aborting/ending bad remote "
+ "transaction. (tx %d.%d, atid %d.%d, abortalways %d, "
+ "endit %d)\n",
+ ubik_currentTrans->tid.epoch,
+ ubik_currentTrans->tid.counter,
+ atid->epoch, atid->counter,
+ abortalways, endit));
+ if (endit) {
udisk_end(ubik_currentTrans);
}
ubik_currentTrans = (struct ubik_trans *)0;
/* otherwise, skip over the data bytes, too */
tpos += ntohl(buffer[2]) + 3 * sizeof(afs_int32);
} else {
- ubik_print("corrupt log opcode (%d) at position %d\n", opcode,
- tpos);
+ ViceLog(0, ("corrupt log opcode (%d) at position %d\n", opcode,
+ tpos));
break; /* corrupt log! */
}
}
code = (*adbase->setlabel) (adbase, 0, &version);
if (code)
return code;
- ubik_print("Successfully replayed log for interrupted "
+ ViceLog(0, ("Successfully replayed log for interrupted "
"transaction; db version is now %ld.%ld\n",
- (long) version.epoch, (long) version.counter);
+ (long) version.epoch, (long) version.counter));
logIsGood = 1;
break; /* all done now */
} else if (opcode == LOGTRUNCATE) {
len -= thisSize;
}
} else {
- ubik_print("corrupt log opcode (%d) at position %d\n",
- opcode, tpos);
+ ViceLog(0, ("corrupt log opcode (%d) at position %d\n",
+ opcode, tpos));
break; /* corrupt log! */
}
}
if (code)
return code;
} else {
- ubik_print("Log read error on pass 2\n");
+ ViceLog(0, ("Log read error on pass 2\n"));
return UBADLOG;
}
}
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);
adbase->version.counter = 0;
(*adbase->setlabel) (adbase, 0, &adbase->version);
}
-#ifdef AFS_PTHREAD_ENV
- CV_BROADCAST(&adbase->version_cond);
-#else
- LWP_NoYieldSignal(&adbase->version);
-#endif
+ UBIK_VERSION_UNLOCK;
}
return 0;
}
{
afs_int32 code;
+ DBHOLD(adbase);
code = ReplayLog(adbase);
if (code)
- return code;
+ goto done;
code = InitializeDB(adbase);
+done:
+ DBRELE(adbase);
return code;
}
void *
urecovery_Interact(void *dummy)
{
- afs_int32 code, tcode;
+ afs_int32 code;
struct ubik_server *bestServer = NULL;
struct ubik_server *ts;
int dbok, doingRPC, now;
int fd = -1;
afs_int32 pass;
+ opr_threadname_set("recovery");
+
/* otherwise, begin interaction */
urecovery_state = 0;
lastProbeTime = 0;
IOMGR_Select(0, 0, 0, 0, &tv);
#endif
- ubik_dprint("recovery running in state %x\n", urecovery_state);
+ 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
*/
if ((now = FT_ApproxTime()) > 30 + lastProbeTime) {
-#ifdef AFS_PTHREAD_ENV
- DBHOLD(ubik_dbase);
-#endif
-
for (ts = ubik_servers, doingRPC = 0; ts; ts = ts->next) {
UBIK_BEACON_LOCK;
if (!ts->up) {
if (code == 0) {
UBIK_BEACON_LOCK;
ts->up = 1;
+ UBIK_BEACON_UNLOCK;
+ DBHOLD(ubik_dbase);
urecovery_state &= ~UBIK_RECFOUNDDB;
+ DBRELE(ubik_dbase);
}
- } else if (!ts->currentDB) {
- urecovery_state &= ~UBIK_RECFOUNDDB;
+ } else {
+ UBIK_BEACON_UNLOCK;
+ DBHOLD(ubik_dbase);
+ if (!ts->currentDB)
+ urecovery_state &= ~UBIK_RECFOUNDDB;
+ DBRELE(ubik_dbase);
}
- UBIK_BEACON_UNLOCK;
}
-#ifdef AFS_PTHREAD_ENV
- DBRELE(ubik_dbase);
-#endif
-
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;
* 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;
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 */
}
}
}
- /* 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;
+
+ 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))
+ 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.
urecovery_state |= UBIK_RECHAVEDB;
} else {
/* we don't have the best version; we should fetch it. */
- DBHOLD(ubik_dbase);
urecovery_AbortAll(ubik_dbase);
/* Rx code to do the Bulk fetch */
file = 0;
offset = 0;
+ UBIK_ADDR_LOCK;
rxcall = rx_NewCall(bestServer->disk_rxcid);
- ubik_print("Ubik: Synchronize database with server %s\n",
- afs_inet_ntoa_r(bestServer->addr[0], hoststr));
+ ViceLog(0, ("Ubik: Synchronize database: receive (via GetFile) "
+ "from server %s begin\n",
+ afs_inet_ntoa_r(bestServer->addr[0], hoststr)));
+ UBIK_ADDR_UNLOCK;
code = StartDISK_GetFile(rxcall, file);
if (code) {
- ubik_dprint("StartDiskGetFile failed=%d\n", code);
+ ViceLog(0, ("StartDiskGetFile failed=%d\n", code));
goto FetchEndCall;
}
nbytes = rx_Read(rxcall, (char *)&length, sizeof(afs_int32));
length = ntohl(length);
if (nbytes != sizeof(afs_int32)) {
- ubik_dprint("Rx-read length error=%d\n", code = BULK_ERROR);
+ ViceLog(0, ("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) {
- ubik_dprint("setlabel io error=%d\n", code);
+ ViceLog(0, ("setlabel io error=%d\n", code));
goto FetchEndCall;
}
snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP",
#endif
nbytes = rx_Read(rxcall, tbuffer, tlen);
if (nbytes != tlen) {
- ubik_dprint("Rx-read bulk error=%d\n", code = BULK_ERROR);
+ ViceLog(0, ("Rx-read bulk error=%d\n", BULK_ERROR));
code = EIO;
close(fd);
goto FetchEndCall;
goto FetchEndCall;
code = EndDISK_GetFile(rxcall, &tversion);
FetchEndCall:
- tcode = rx_EndCall(rxcall, code);
- if (!code)
- code = tcode;
+ 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->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":"",
* 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_print("Ubik: Synchronize database failed (error = %d)\n",
- code);
+ UBIK_VERSION_UNLOCK;
+ ViceLog(0,
+ ("Ubik: Synchronize database: receive (via GetFile) "
+ "from server %s failed (error = %d)\n",
+ afs_inet_ntoa_r(bestServer->addr[0], hoststr), code));
} else {
- ubik_print("Ubik: Synchronize database completed\n");
+ ViceLog(0,
+ ("Ubik: Synchronize database: receive (via GetFile) "
+ "from server %s complete, version: %d.%d\n",
+ afs_inet_ntoa_r(bestServer->addr[0], hoststr),
+ ubik_dbase->version.epoch, ubik_dbase->version.counter));
+
urecovery_state |= UBIK_RECHAVEDB;
}
udisk_Invalidate(ubik_dbase, 0); /* data has changed */
-#ifdef AFS_PTHREAD_ENV
- CV_BROADCAST(&ubik_dbase->version_cond);
-#else
- LWP_NoYieldSignal(&ubik_dbase->version);
-#endif
- DBRELE(ubik_dbase);
}
- if (!(urecovery_state & UBIK_RECHAVEDB))
+ 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.
* database and overwrite this one.
*/
if (ubik_dbase->version.epoch == 1) {
- DBHOLD(ubik_dbase);
urecovery_AbortAll(ubik_dbase);
- ubik_epochTime = 2;
- ubik_dbase->version.epoch = ubik_epochTime;
+ 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
- CV_BROADCAST(&ubik_dbase->version_cond);
-#else
- LWP_NoYieldSignal(&ubik_dbase->version);
-#endif
- DBRELE(ubik_dbase);
}
/* Check the other sites and send the database to them if they
/* now propagate out new version to everyone else */
dbok = 1; /* start off assuming they all worked */
- DBHOLD(ubik_dbase);
/*
* 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
if (ubik_dbase->flags & DBWRITING) {
struct timeval tv;
int safety = 0;
- tv.tv_sec = 0;
- tv.tv_usec = 50000;
+ 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
- tv.tv_usec += 10000;
+ 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;
- ubik_dprint("recovery cannot send version to %s\n",
- afs_inet_ntoa_r(inAddr.s_addr, hoststr));
+ /* It would be nice to have this message at loglevel
+ * 0 as well, but it will log once every 4s for each
+ * down server while in this recovery state. This
+ * should only be changed to loglevel 0 if it is
+ * also rate-limited.
+ */
+ ViceLog(5, ("recovery cannot send version to %s\n",
+ afs_inet_ntoa_r(inAddr.s_addr, hoststr)));
dbok = 0;
continue;
}
UBIK_BEACON_UNLOCK;
- ubik_dprint("recovery sending version to %s\n",
- afs_inet_ntoa_r(inAddr.s_addr, hoststr));
+
if (vcmp(ts->version, ubik_dbase->version) != 0) {
- ubik_dprint("recovery stating local database\n");
+ ViceLog(0, ("Synchronize database: send (via SendFile) "
+ "to server %s begin\n",
+ afs_inet_ntoa_r(inAddr.s_addr, hoststr)));
/* 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) {
- ubik_dprint("StartDiskSendFile failed=%d\n",
- code);
+ ViceLog(0, ("StartDiskSendFile failed=%d\n",
+ code));
goto StoreEndCall;
}
while (length > 0) {
(*ubik_dbase->read) (ubik_dbase, file,
tbuffer, offset, tlen);
if (nbytes != tlen) {
- ubik_dprint("Local disk read error=%d\n",
- code = UIOERROR);
+ code = UIOERROR;
+ ViceLog(0, ("Local disk read error=%d\n", code));
goto StoreEndCall;
}
nbytes = rx_Write(rxcall, tbuffer, tlen);
if (nbytes != tlen) {
- ubik_dprint("Rx-write bulk error=%d\n", code =
- BULK_ERROR);
+ code = BULK_ERROR;
+ ViceLog(0, ("Rx-write bulk error=%d\n", code));
goto StoreEndCall;
}
offset += tlen;
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
+ ViceLog(0,
+ ("Ubik: Synchronize database: send (via SendFile) "
+ "to server %s complete, version: %d.%d\n",
+ afs_inet_ntoa_r(inAddr.s_addr, hoststr),
+ ts->version.epoch, ts->version.counter));
+
+ } else {
dbok = 0;
+ ViceLog(0,
+ ("Ubik: Synchronize database: send (via SendFile) "
+ "to server %s failed (error = %d)\n",
+ afs_inet_ntoa_r(inAddr.s_addr, hoststr), code));
+ }
} else {
/* mark file up to date */
ts->currentDB = 1;
}
}
- DBRELE(ubik_dbase);
if (dbok)
urecovery_state |= UBIK_RECSENTDB;
}
+ DBRELE(ubik_dbase);
}
- return NULL;
+ AFS_UNREACHED(return(NULL));
}
/*!
{
struct rx_connection *conns[UBIK_MAX_INTERFACE_ADDR];
struct rx_connection *connSuccess = 0;
- int i, j, success_i = -1;
+ int i, nconns, success_i = -1;
afs_uint32 addr;
char buffer[32];
char hoststr[16];
- extern afs_int32 ubikSecIndex;
- extern struct rx_securityClass *ubikSecClass;
- for (i = 0; (addr = server->addr[i]) && (i < UBIK_MAX_INTERFACE_ADDR);
+ 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,
- ubikSecClass, ubikSecIndex);
+ addr_globals.ubikSecClass, addr_globals.ubikSecIndex);
/* user requirement to use only the primary interface */
if (ubikPrimaryAddrOnly) {
break;
}
}
- osi_Assert(i); /* at least one interface address for this server */
-
-#ifdef AFS_PTHREAD_ENV
- DBRELE(ubik_dbase);
-#endif
+ UBIK_ADDR_UNLOCK;
+ nconns = i;
+ opr_Assert(nconns); /* at least one interface address for this server */
- multi_Rx(conns, i) {
+ multi_Rx(conns, nconns) {
multi_DISK_Probe();
if (!multi_error) { /* first success */
success_i = multi_i;
}
} multi_End_Ignore;
-#ifdef AFS_PTHREAD_ENV
- DBHOLD(ubik_dbase);
-#endif
-
if (success_i >= 0) {
+ UBIK_ADDR_LOCK;
addr = server->addr[success_i]; /* successful interface addr */
if (server->disk_rxcid) /* destroy existing conn */
/* make new connections */
server->disk_rxcid = conns[success_i];
server->vote_rxcid = rx_NewConnection(addr, ubik_callPortal,
- VOTE_SERVICE_ID, ubikSecClass,
- ubikSecIndex);
+ VOTE_SERVICE_ID, addr_globals.ubikSecClass,
+ addr_globals.ubikSecIndex);
connSuccess = conns[success_i];
strcpy(buffer, afs_inet_ntoa_r(server->addr[0], hoststr));
- ubik_print("ubik:server %s is back up: will be contacted through %s\n",
- buffer, afs_inet_ntoa_r(addr, 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 (j = 0; j < i; j++)
- if (conns[j] != connSuccess)
- rx_DestroyConnection(conns[j]);
+ for (i = 0; i < nconns; i++)
+ if (conns[i] != connSuccess)
+ rx_DestroyConnection(conns[i]);
if (!connSuccess)
- ubik_dprint("ubik:server %s still down\n",
- afs_inet_ntoa_r(server->addr[0], hoststr));
+ ViceLog(5, ("ubik:server %s still down\n",
+ afs_inet_ntoa_r(server->addr[0], hoststr)));
if (connSuccess)
return 0; /* success */