osi_rwlock_t cm_serverLock;
osi_rwlock_t cm_syscfgLock;
-cm_server_t *cm_allServersp;
+cm_server_t *cm_serversAllFirstp = NULL;
+cm_server_t *cm_serversAllLastp = NULL;
+
afs_uint32 cm_numFileServers = 0;
afs_uint32 cm_numVldbServers = 0;
cm_server_t *tsp;
lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp = tsp->allNextp) {
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
cm_GetServerNoLock(tsp);
lock_ReleaseRead(&cm_serverLock);
cm_ForceNewConnections(tsp);
lock_ReleaseRead(&cm_serverLock);
}
+void
+cm_ServerClearRPCStats(void) {
+ cm_server_t *tsp;
+ afs_uint16 port;
+
+ lock_ObtainRead(&cm_serverLock);
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
+ switch (tsp->type) {
+ case CM_SERVER_VLDB:
+ port = htons(7003);
+ rx_ClearPeerRPCStats(opcode_VL_ProbeServer>>32, tsp->addr.sin_addr.s_addr, port);
+ break;
+ case CM_SERVER_FILE:
+ port = htons(7000);
+ rx_ClearPeerRPCStats(opcode_RXAFS_GetCapabilities>>32, tsp->addr.sin_addr.s_addr, port);
+ rx_ClearPeerRPCStats(opcode_RXAFS_GetTime>>32, tsp->addr.sin_addr.s_addr, port);
+ break;
+ }
+ }
+ lock_ReleaseRead(&cm_serverLock);
+}
+
/*
* lock_ObtainMutex must be held prior to calling
* this function.
{
afs_int32 code = 0; /* start with "success" */
struct rx_debugPeer tpeer;
+ struct rx_peer * rxPeer;
afs_uint16 port;
- afs_uint16 newRank;
+ afs_uint64 newRank;
+ afs_uint64 perfRank = 0;
+ afs_uint64 rtt = 0;
+ double log_rtt;
+
+ int isDown = (tsp->flags & CM_SERVERFLAG_DOWN);
+ void *peerRpcStats = NULL;
+ afs_uint64 opcode = 0;
switch(tsp->type) {
case CM_SERVER_VLDB:
port = htons(7003);
+ opcode = opcode_VL_ProbeServer;
break;
case CM_SERVER_FILE:
port = htons(7000);
+ opcode = opcode_RXAFS_GetCapabilities;
break;
default:
return -1;
}
- code = rx_GetLocalPeers(tsp->addr.sin_addr.s_addr, port, &tpeer);
+ cm_SetServerIPRank(tsp);
- /*check if rx_GetLocalPeers succeeded and if there is data for tsp */
- if(code == 0 && (tpeer.rtt == 0 && tpeer.rtt_dev == 0))
- code = -1;
+ if (isDown) {
+ newRank = 0xFFFF;
+ } else {
+ /*
+ * There are three potential components to the ranking:
+ * 1. Any administrative set preference whether it be
+ * via "fs setserverprefs", registry or dns.
+ *
+ * 2. Network subnet mask comparison.
+ *
+ * 3. Performance data.
+ *
+ * If there is an administrative rank, that is the
+ * the primary factor. If not the primary factor
+ * is the network ranking.
+ */
+
+ code = rx_GetLocalPeers(tsp->addr.sin_addr.s_addr, port, &tpeer);
+ if (code == 0) {
+ peerRpcStats = rx_CopyPeerRPCStats(opcode, tsp->addr.sin_addr.s_addr, port);
+ if (peerRpcStats == NULL && tsp->type == CM_SERVER_FILE)
+ peerRpcStats = rx_CopyPeerRPCStats(opcode_RXAFS_GetTime, tsp->addr.sin_addr.s_addr, port);
+ if (peerRpcStats) {
+ afs_uint64 execTimeSum = _8THMSEC(RPCOpStat_ExecTimeSum(peerRpcStats));
+ afs_uint64 queueTimeSum = _8THMSEC(RPCOpStat_QTimeSum(peerRpcStats));
+ afs_uint64 numCalls = RPCOpStat_NumCalls(peerRpcStats);
+
+ if (numCalls > 0)
+ rtt = (execTimeSum - queueTimeSum) / numCalls;
+
+ rx_ReleaseRPCStats(peerRpcStats);
+ }
- if(code == 0) {
- if((tsp->flags & CM_SERVERFLAG_PREF_SET))
- newRank = tsp->adminRank +
- ((int)(623 * log(tpeer.rtt) / 10) * 10 + 5);
- else /* rank has not been set by admin, derive rank from rtt */
- newRank = (int)(7200 * log(tpeer.rtt) / 5000) * 5000 + 5000;
+ if (rtt == 0 && tpeer.rtt) {
+ /* rtt is ms/8 */
+ rtt = tpeer.rtt;
+ }
- newRank += (rand() & 0x000f); /* randomize */
+ if (rtt > 0) {
+ log_rtt = log(rtt);
+ perfRank += (6000 * log_rtt / 5000) * 5000;
- if (abs(newRank - tsp->ipRank) > 0xf) {
- tsp->ipRank = newRank;
+ if (tsp->type == CM_SERVER_FILE) {
+ /* give an edge to servers with high congestion windows */
+ perfRank -= (tpeer.cwind - 1)* 15;
+ }
+ }
+ }
+
+ if (tsp->adminRank) {
+ newRank = tsp->adminRank * 0.8;
+ newRank += tsp->ipRank * 0.2;
+ } else {
+ newRank = tsp->ipRank;
+ }
+ if (perfRank) {
+ newRank *= 0.9;
+ newRank += perfRank * 0.1;
+ }
+ newRank += (rand() & 0x000f); /* randomize */
+
+ if (newRank > 0xFFFF)
+ osi_Log1(afsd_logp, "new server rank %I64u exceeds 0xFFFF", newRank);
+
+ /*
+ * If the ranking changes by more than the randomization
+ * factor, update the server reference lists.
+ */
+ if (abs(newRank - tsp->activeRank) > 0xf) {
+ tsp->activeRank = newRank;
lock_ReleaseMutex(&tsp->mx);
switch (tsp->type) {
return code;
}
+static void
+cm_MarkServerDown(cm_server_t *tsp, afs_int32 code, int wasDown)
+{
+
+ /* mark server as down */
+ if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_DOWN);
+ tsp->downTime = time(NULL);
+ }
+ if (code != VRESTARTING) {
+ lock_ReleaseMutex(&tsp->mx);
+ cm_ForceNewConnections(tsp);
+ lock_ObtainMutex(&tsp->mx);
+ }
+ /* Now update the volume status if necessary */
+ if (!wasDown) {
+ if (tsp->type == CM_SERVER_FILE) {
+ cm_server_vols_t * tsrvp;
+ cm_volume_t * volp;
+ int i;
+ cm_req_t req;
+
+ for (tsrvp = tsp->vols; tsrvp; tsrvp = tsrvp->nextp) {
+ for (i=0; i<NUM_SERVER_VOLS; i++) {
+ if (tsrvp->ids[i] != 0) {
+ cm_InitReq(&req);
+
+ lock_ReleaseMutex(&tsp->mx);
+ code = cm_FindVolumeByID(tsp->cellp, tsrvp->ids[i],
+ cm_rootUserp, &req,
+ CM_GETVOL_FLAG_NO_LRU_UPDATE,
+ &volp);
+ lock_ObtainMutex(&tsp->mx);
+ if (code == 0) {
+ cm_UpdateVolumeStatus(volp, tsrvp->ids[i]);
+ cm_PutVolume(volp);
+ }
+ }
+ }
+ }
+ }
+ cm_RankServer(tsp);
+ }
+}
+
void
cm_PingServer(cm_server_t *tsp)
{
lock_ObtainMutex(&tsp->mx);
tsp->waitCount--;
if (tsp->waitCount == 0)
- tsp->flags &= ~CM_SERVERFLAG_PINGING;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_PINGING);
else
osi_Wakeup((LONG_PTR)tsp);
lock_ReleaseMutex(&tsp->mx);
return;
}
- tsp->flags |= CM_SERVERFLAG_PINGING;
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_PINGING);
wasDown = tsp->flags & CM_SERVERFLAG_DOWN;
afs_inet_ntoa_r(tsp->addr.sin_addr.S_un.S_addr, hoststr);
lock_ReleaseMutex(&tsp->mx);
- code = cm_ConnByServer(tsp, cm_rootUserp, &connp);
+ code = cm_ConnByServer(tsp, cm_rootUserp, FALSE, &connp);
if (code == 0) {
/* now call the appropriate ping call. Drop the timeout if
* the server is known to be down, so that we don't waste a
rxconnp = cm_GetRxConn(connp);
if (wasDown)
- rx_SetConnDeadTime(rxconnp, 10);
+ rx_SetConnHardDeadTime(rxconnp, 10);
if (tsp->type == CM_SERVER_VLDB) {
code = VL_ProbeServer(rxconnp);
}
code = RXAFS_GetCapabilities(rxconnp, &caps);
}
if (wasDown)
- rx_SetConnDeadTime(rxconnp, ConnDeadtimeout);
+ rx_SetConnHardDeadTime(rxconnp, HardDeadtimeout);
rx_PutConnection(rxconnp);
cm_PutConn(connp);
} /* got an unauthenticated connection to this server */
lock_ObtainMutex(&tsp->mx);
- if (code >= 0 || code == RXGEN_OPCODE) {
+ if (code >= 0 || code == RXGEN_OPCODE || code == RX_CALL_BUSY) {
/* mark server as up */
- tsp->flags &= ~CM_SERVERFLAG_DOWN;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_DOWN);
tsp->downTime = 0;
/* we currently handle 32-bits of capabilities */
- if (code != RXGEN_OPCODE && caps.Capabilities_len > 0) {
+ if (code != RXGEN_OPCODE && code != RX_CALL_BUSY &&
+ caps.Capabilities_len > 0) {
tsp->capabilities = caps.Capabilities_val[0];
xdr_free((xdrproc_t) xdr_Capabilities, &caps);
caps.Capabilities_len = 0;
}
}
}
+ cm_RankServer(tsp);
}
} else {
- /* mark server as down */
- if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
- tsp->flags |= CM_SERVERFLAG_DOWN;
- tsp->downTime = time(NULL);
- }
- if (code != VRESTARTING) {
- lock_ReleaseMutex(&tsp->mx);
- cm_ForceNewConnections(tsp);
- lock_ObtainMutex(&tsp->mx);
- }
+ cm_MarkServerDown(tsp, code, wasDown);
+
osi_Log3(afsd_logp, "cm_PingServer server %s (%s) is down with caps 0x%x",
osi_LogSaveString(afsd_logp, hoststr),
tsp->type == CM_SERVER_VLDB ? "vldb" : "file",
tsp->capabilities);
-
- /* Now update the volume status if necessary */
- if (!wasDown) {
- cm_server_vols_t * tsrvp;
- cm_volume_t * volp;
- int i;
-
- for (tsrvp = tsp->vols; tsrvp; tsrvp = tsrvp->nextp) {
- for (i=0; i<NUM_SERVER_VOLS; i++) {
- if (tsrvp->ids[i] != 0) {
- cm_InitReq(&req);
-
- lock_ReleaseMutex(&tsp->mx);
- code = cm_FindVolumeByID(tsp->cellp, tsrvp->ids[i], cm_rootUserp,
- &req, CM_GETVOL_FLAG_NO_LRU_UPDATE, &volp);
- lock_ObtainMutex(&tsp->mx);
- if (code == 0) {
- cm_UpdateVolumeStatus(volp, tsrvp->ids[i]);
- cm_PutVolume(volp);
- }
- }
- }
- }
- }
}
if (tsp->waitCount == 0)
- tsp->flags &= ~CM_SERVERFLAG_PINGING;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_PINGING);
else
osi_Wakeup((LONG_PTR)tsp);
lock_ReleaseMutex(&tsp->mx);
cm_server_t * tsp;
lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp = tsp->allNextp) {
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
cm_GetServerNoLock(tsp);
lock_ReleaseRead(&cm_serverLock);
int isVLDB;
lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp = tsp->allNextp) {
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
cm_GetServerNoLock(tsp);
lock_ReleaseRead(&cm_serverLock);
cm_server_t ** serversp, *tsp;
afs_uint32 isDown, wasDown;
afs_uint32 code;
- time_t start, *deltas;
+ time_t start;
char hoststr[16];
cm_InitReq(&req);
conns = (cm_conn_t **)malloc(maxconns * sizeof(cm_conn_t *));
rxconns = (struct rx_connection **)malloc(maxconns * sizeof(struct rx_connection *));
conntimer = (afs_int32 *)malloc(maxconns * sizeof (afs_int32));
- deltas = (time_t *)malloc(maxconns * sizeof (time_t));
results = (afs_int32 *)malloc(maxconns * sizeof (afs_int32));
serversp = (cm_server_t **)malloc(maxconns * sizeof(cm_server_t *));
caps = (Capabilities *)malloc(maxconns * sizeof(Capabilities));
!(flags & (CM_FLAG_CHECKFILESERVERS|CM_FLAG_CHECKVLDBSERVERS)))
{
lock_ObtainRead(&cm_serverLock);
- for (nconns=0, tsp = cm_allServersp; tsp && nconns < maxconns; tsp = tsp->allNextp) {
+ for (nconns=0, tsp = cm_serversAllFirstp;
+ tsp != NULL && nconns < maxconns;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
if (tsp->type != CM_SERVER_FILE ||
tsp->cellp == NULL || /* SetPref only */
cellp && cellp != tsp->cellp)
continue;
}
- tsp->flags |= CM_SERVERFLAG_PINGING;
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_PINGING);
lock_ReleaseMutex(&tsp->mx);
serversp[nconns] = tsp;
- code = cm_ConnByServer(tsp, cm_rootUserp, &conns[nconns]);
+ code = cm_ConnByServer(tsp, cm_rootUserp, FALSE, &conns[nconns]);
if (code) {
lock_ObtainRead(&cm_serverLock);
cm_PutServerNoLock(tsp);
lock_ObtainRead(&cm_serverLock);
rxconns[nconns] = cm_GetRxConn(conns[nconns]);
if (conntimer[nconns] = (isDown ? 1 : 0))
- rx_SetConnDeadTime(rxconns[nconns], 10);
+ rx_SetConnHardDeadTime(rxconns[nconns], 10);
nconns++;
}
/* Process results of servers that support RXAFS_GetCapabilities */
for (i=0; i<nconns; i++) {
if (conntimer[i])
- rx_SetConnDeadTime(rxconns[i], ConnDeadtimeout);
+ rx_SetConnHardDeadTime(rxconns[i], HardDeadtimeout);
rx_PutConnection(rxconns[i]);
cm_PutConn(conns[i]);
lock_ObtainMutex(&tsp->mx);
wasDown = tsp->flags & CM_SERVERFLAG_DOWN;
- if (results[i] >= 0 || results[i] == RXGEN_OPCODE) {
+ if (results[i] >= 0 || results[i] == RXGEN_OPCODE ||
+ results[i] == RX_CALL_BUSY) {
/* mark server as up */
- tsp->flags &= ~CM_SERVERFLAG_DOWN;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_DOWN);
tsp->downTime = 0;
/* we currently handle 32-bits of capabilities */
- if (results[i] != RXGEN_OPCODE && caps[i].Capabilities_len > 0) {
+ if (results[i] != RXGEN_OPCODE && results[i] != RX_CALL_BUSY &&
+ caps[i].Capabilities_len > 0) {
tsp->capabilities = caps[i].Capabilities_val[0];
xdr_free((xdrproc_t) xdr_Capabilities, &caps[i]);
caps[i].Capabilities_len = 0;
}
}
}
+ cm_RankServer(tsp);
}
} else {
- /* mark server as down */
- if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
- tsp->flags |= CM_SERVERFLAG_DOWN;
- tsp->downTime = time(NULL);
- }
- if (code != VRESTARTING) {
- lock_ReleaseMutex(&tsp->mx);
- cm_ForceNewConnections(tsp);
- lock_ObtainMutex(&tsp->mx);
- }
- afs_inet_ntoa_r(tsp->addr.sin_addr.S_un.S_addr, hoststr);
+ cm_MarkServerDown(tsp, results[i], wasDown);
+
+ afs_inet_ntoa_r(tsp->addr.sin_addr.S_un.S_addr, hoststr);
osi_Log3(afsd_logp, "cm_MultiPingServer server %s (%s) is down with caps 0x%x",
osi_LogSaveString(afsd_logp, hoststr),
tsp->type == CM_SERVER_VLDB ? "vldb" : "file",
tsp->capabilities);
-
- /* Now update the volume status if necessary */
- if (!wasDown) {
- cm_server_vols_t * tsrvp;
- cm_volume_t * volp;
- int i;
-
- for (tsrvp = tsp->vols; tsrvp; tsrvp = tsrvp->nextp) {
- for (i=0; i<NUM_SERVER_VOLS; i++) {
- if (tsrvp->ids[i] != 0) {
- cm_InitReq(&req);
-
- lock_ReleaseMutex(&tsp->mx);
- code = cm_FindVolumeByID(tsp->cellp, tsrvp->ids[i], cm_rootUserp,
- &req, CM_GETVOL_FLAG_NO_LRU_UPDATE, &volp);
- lock_ObtainMutex(&tsp->mx);
- if (code == 0) {
- cm_UpdateVolumeStatus(volp, tsrvp->ids[i]);
- cm_PutVolume(volp);
- }
- }
- }
- }
- }
}
if (tsp->waitCount == 0)
- tsp->flags &= ~CM_SERVERFLAG_PINGING;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_PINGING);
else
osi_Wakeup((LONG_PTR)tsp);
!(flags & (CM_FLAG_CHECKFILESERVERS|CM_FLAG_CHECKVLDBSERVERS)))
{
lock_ObtainRead(&cm_serverLock);
- for (nconns=0, tsp = cm_allServersp; tsp && nconns < maxconns; tsp = tsp->allNextp) {
+ for (nconns=0, tsp = cm_serversAllFirstp;
+ tsp != NULL && nconns < maxconns;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
if (tsp->type != CM_SERVER_VLDB ||
tsp->cellp == NULL || /* SetPref only */
cellp && cellp != tsp->cellp)
continue;
}
- tsp->flags |= CM_SERVERFLAG_PINGING;
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_PINGING);
lock_ReleaseMutex(&tsp->mx);
serversp[nconns] = tsp;
- code = cm_ConnByServer(tsp, cm_rootUserp, &conns[nconns]);
+ code = cm_ConnByServer(tsp, cm_rootUserp, FALSE, &conns[nconns]);
if (code) {
lock_ObtainRead(&cm_serverLock);
cm_PutServerNoLock(tsp);
rxconns[nconns] = cm_GetRxConn(conns[nconns]);
conntimer[nconns] = (isDown ? 1 : 0);
if (isDown)
- rx_SetConnDeadTime(rxconns[nconns], 10);
+ rx_SetConnHardDeadTime(rxconns[nconns], 10);
nconns++;
}
/* Process results of servers that support VL_ProbeServer */
for (i=0; i<nconns; i++) {
if (conntimer[i])
- rx_SetConnDeadTime(rxconns[i], ConnDeadtimeout);
+ rx_SetConnHardDeadTime(rxconns[i], HardDeadtimeout);
rx_PutConnection(rxconns[i]);
cm_PutConn(conns[i]);
lock_ObtainMutex(&tsp->mx);
wasDown = tsp->flags & CM_SERVERFLAG_DOWN;
- if (results[i] >= 0) {
+ if (results[i] >= 0 || results[i] == RX_CALL_BUSY) {
/* mark server as up */
- tsp->flags &= ~CM_SERVERFLAG_DOWN;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_DOWN);
tsp->downTime = 0;
tsp->capabilities = 0;
osi_LogSaveString(afsd_logp, hoststr),
tsp->type == CM_SERVER_VLDB ? "vldb" : "file",
tsp->capabilities);
+ if (wasDown)
+ cm_RankServer(tsp);
} else {
- /* mark server as down */
- if (!(tsp->flags & CM_SERVERFLAG_DOWN)) {
- tsp->flags |= CM_SERVERFLAG_DOWN;
- tsp->downTime = time(NULL);
- }
- if (code != VRESTARTING) {
- lock_ReleaseMutex(&tsp->mx);
- cm_ForceNewConnections(tsp);
- lock_ObtainMutex(&tsp->mx);
- }
- afs_inet_ntoa_r(tsp->addr.sin_addr.S_un.S_addr, hoststr);
+ cm_MarkServerDown(tsp, results[i], wasDown);
+
+ afs_inet_ntoa_r(tsp->addr.sin_addr.S_un.S_addr, hoststr);
osi_Log3(afsd_logp, "cm_MultiPingServer server %s (%s) is down with caps 0x%x",
osi_LogSaveString(afsd_logp, hoststr),
tsp->type == CM_SERVER_VLDB ? "vldb" : "file",
}
if (tsp->waitCount == 0)
- tsp->flags &= ~CM_SERVERFLAG_PINGING;
+ _InterlockedAnd(&tsp->flags, ~CM_SERVERFLAG_PINGING);
else
osi_Wakeup((LONG_PTR)tsp);
free(conns);
free(rxconns);
free(conntimer);
- free(deltas);
free(results);
free(serversp);
free(caps);
{
lock_ObtainMutex(&serverp->mx);
if (no64bit)
- serverp->flags |= CM_SERVERFLAG_NO64BIT;
+ _InterlockedOr(&serverp->flags, CM_SERVERFLAG_NO64BIT);
else
- serverp->flags &= ~CM_SERVERFLAG_NO64BIT;
+ _InterlockedAnd(&serverp->flags, ~CM_SERVERFLAG_NO64BIT);
lock_ReleaseMutex(&serverp->mx);
}
{
lock_ObtainMutex(&serverp->mx);
if (no)
- serverp->flags |= CM_SERVERFLAG_NOINLINEBULK;
+ _InterlockedOr(&serverp->flags, CM_SERVERFLAG_NOINLINEBULK);
else
- serverp->flags &= ~CM_SERVERFLAG_NOINLINEBULK;
+ _InterlockedAnd(&serverp->flags, ~CM_SERVERFLAG_NOINLINEBULK);
lock_ReleaseMutex(&serverp->mx);
}
-void cm_SetServerPrefs(cm_server_t * serverp)
+afs_int32 cm_UpdateIFInfo(void)
+{
+ afs_int32 code;
+ /* get network related info */
+ cm_noIPAddr = CM_MAXINTERFACE_ADDR;
+ code = syscfg_GetIFInfo(&cm_noIPAddr,
+ cm_IPAddr, cm_SubnetMask,
+ cm_NetMtu, cm_NetFlags);
+ cm_LanAdapterChangeDetected = 0;
+ return code;
+}
+
+void cm_SetServerIPRank(cm_server_t * serverp)
{
unsigned long serverAddr; /* in host byte order */
unsigned long myAddr, myNet, mySubnet;/* in host byte order */
unsigned long netMask;
int i;
- long code;
- int writeLock = 0;
+ afs_int32 code;
lock_ObtainRead(&cm_syscfgLock);
if (cm_LanAdapterChangeDetected) {
lock_ConvertRToW(&cm_syscfgLock);
- writeLock = 1;
if (cm_LanAdapterChangeDetected) {
- /* get network related info */
- cm_noIPAddr = CM_MAXINTERFACE_ADDR;
- code = syscfg_GetIFInfo(&cm_noIPAddr,
- cm_IPAddr, cm_SubnetMask,
- cm_NetMtu, cm_NetFlags);
- cm_LanAdapterChangeDetected = 0;
- }
+ code = cm_UpdateIFInfo();
+ }
lock_ConvertWToR(&cm_syscfgLock);
}
{
if ( (serverAddr & cm_SubnetMask[i]) == mySubnet)
{
- if ( serverAddr == myAddr )
+ if ( serverAddr == myAddr ) {
serverp->ipRank = min(serverp->ipRank,
CM_IPRANK_TOP);/* same machine */
- else serverp->ipRank = min(serverp->ipRank,
- CM_IPRANK_HI); /* same subnet */
- }
- else serverp->ipRank = min(serverp->ipRank,CM_IPRANK_MED);
- /* same net */
+ } else {
+ serverp->ipRank = min(serverp->ipRank,
+ CM_IPRANK_HI); /* same subnet */
+ }
+ } else {
+ serverp->ipRank = min(serverp->ipRank, CM_IPRANK_MED); /* same net */
+ }
}
} /* and of for loop */
-
- /* random between 0..15*/
- serverp->ipRank += (rand() % 0x000f);
lock_ReleaseRead(&cm_syscfgLock);
}
cm_server_t *cm_NewServer(struct sockaddr_in *socketp, int type, cm_cell_t *cellp, afsUUID *uuidp, afs_uint32 flags) {
cm_server_t *tsp;
+ char hoststr[16];
osi_assertx(socketp->sin_family == AF_INET, "unexpected socket family");
+ lock_ObtainWrite(&cm_serverLock); /* get server lock */
+ tsp = cm_FindServer(socketp, type, TRUE);
+ if (tsp) {
+ /* we might have found a server created by set server prefs */
+ if (uuidp && !afs_uuid_is_nil(uuidp) &&
+ !(tsp->flags & CM_SERVERFLAG_UUID))
+ {
+ tsp->uuid = *uuidp;
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_UUID);
+ }
+
+ if (cellp != NULL && tsp->cellp == NULL) {
+ tsp->cellp = cellp;
+ afs_inet_ntoa_r(tsp->addr.sin_addr.s_addr, hoststr);
+ osi_Log3(afsd_logp, "cm_NewServer assigning server %s to cell (%u) %s",
+ osi_LogSaveString(afsd_logp,hoststr),
+ cellp->cellID,
+ osi_LogSaveString(afsd_logp,cellp->name));
+ }
+ else if (tsp->cellp != cellp) {
+ afs_inet_ntoa_r(tsp->addr.sin_addr.s_addr, hoststr);
+ osi_Log5(afsd_logp,
+ "cm_NewServer found a server %s associated with two cells (%u) %s and (%u) %s",
+ osi_LogSaveString(afsd_logp,hoststr),
+ tsp->cellp->cellID,
+ osi_LogSaveString(afsd_logp,tsp->cellp->name),
+ cellp->cellID,
+ osi_LogSaveString(afsd_logp,cellp->name));
+ }
+ lock_ReleaseWrite(&cm_serverLock);
+ return tsp;
+ }
+
tsp = malloc(sizeof(*tsp));
if (tsp) {
memset(tsp, 0, sizeof(*tsp));
tsp->type = type;
- tsp->cellp = cellp;
if (uuidp && !afs_uuid_is_nil(uuidp)) {
tsp->uuid = *uuidp;
- tsp->flags |= CM_SERVERFLAG_UUID;
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_UUID);
}
tsp->refCount = 1;
lock_InitializeMutex(&tsp->mx, "cm_server_t mutex", LOCK_HIERARCHY_SERVER);
tsp->addr = *socketp;
- cm_SetServerPrefs(tsp);
-
- lock_ObtainWrite(&cm_serverLock); /* get server lock */
- tsp->allNextp = cm_allServersp;
- cm_allServersp = tsp;
+ osi_QAddH((osi_queue_t **)&cm_serversAllFirstp,
+ (osi_queue_t **)&cm_serversAllLastp, &tsp->allq);
switch (type) {
case CM_SERVER_VLDB:
break;
}
- lock_ReleaseWrite(&cm_serverLock); /* release server lock */
+ if (cellp != NULL) {
+ tsp->cellp = cellp;
+ afs_inet_ntoa_r(tsp->addr.sin_addr.s_addr, hoststr);
+ osi_Log3(afsd_logp, "cm_NewServer new server %s in cell (%u) %s",
+ osi_LogSaveString(afsd_logp,hoststr),
+ cellp->cellID,
+ osi_LogSaveString(afsd_logp,cellp->name));
+ }
+ }
+ lock_ReleaseWrite(&cm_serverLock); /* release server lock */
+
+ if (tsp) {
+ if (!(flags & CM_FLAG_NOPROBE)) {
+ _InterlockedOr(&tsp->flags, CM_SERVERFLAG_DOWN); /* assume down; ping will mark up if available */
+ lock_ObtainMutex(&tsp->mx);
+ cm_RankServer(tsp);
+ lock_ReleaseMutex(&tsp->mx);
+ cm_PingServer(tsp); /* Obtain Capabilities and check up/down state */
+ } else {
+ pthread_t phandle;
+ pthread_attr_t tattr;
+ int pstatus;
+
+ /* Probe the server in the background to determine if it is up or down */
+ pthread_attr_init(&tattr);
+ pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED);
+
+ lock_ObtainMutex(&tsp->mx);
+ cm_RankServer(tsp);
+ lock_ReleaseMutex(&tsp->mx);
+ pstatus = pthread_create(&phandle, &tattr, cm_PingServer, tsp);
- if ( !(flags & CM_FLAG_NOPROBE) ) {
- tsp->flags |= CM_SERVERFLAG_DOWN; /* assume down; ping will mark up if available */
- cm_PingServer(tsp); /* Obtain Capabilities and check up/down state */
+ pthread_attr_destroy(&tattr);
}
}
return tsp;
}
cm_server_t *
-cm_FindServerByIP(afs_uint32 ipaddr, unsigned short port, int type)
+cm_FindServerByIP(afs_uint32 ipaddr, unsigned short port, int type, int locked)
{
cm_server_t *tsp;
- lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp = tsp->allNextp) {
+ if (!locked)
+ lock_ObtainRead(&cm_serverLock);
+
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
if (tsp->type == type &&
tsp->addr.sin_addr.S_un.S_addr == ipaddr &&
(tsp->addr.sin_port == port || tsp->addr.sin_port == 0))
if (tsp)
cm_GetServerNoLock(tsp);
- lock_ReleaseRead(&cm_serverLock);
+ if (!locked)
+ lock_ReleaseRead(&cm_serverLock);
return tsp;
}
cm_server_t *
-cm_FindServerByUuid(afsUUID *serverUuid, int type)
+cm_FindServerByUuid(afsUUID *serverUuid, int type, int locked)
{
cm_server_t *tsp;
- lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp = tsp->allNextp) {
- if (tsp->type == type && !afs_uuid_equal(&tsp->uuid, serverUuid))
+ if (!locked)
+ lock_ObtainRead(&cm_serverLock);
+
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq)) {
+ if (tsp->type == type && afs_uuid_equal(&tsp->uuid, serverUuid))
break;
}
if (tsp)
cm_GetServerNoLock(tsp);
- lock_ReleaseRead(&cm_serverLock);
+ if (!locked)
+ lock_ReleaseRead(&cm_serverLock);
return tsp;
}
/* find a server based on its properties */
-cm_server_t *cm_FindServer(struct sockaddr_in *addrp, int type)
+cm_server_t *cm_FindServer(struct sockaddr_in *addrp, int type, int locked)
{
- cm_server_t *tsp;
-
osi_assertx(addrp->sin_family == AF_INET, "unexpected socket value");
- lock_ObtainRead(&cm_serverLock);
- for (tsp = cm_allServersp; tsp; tsp=tsp->allNextp) {
- if (tsp->type == type &&
- tsp->addr.sin_addr.s_addr == addrp->sin_addr.s_addr &&
- (tsp->addr.sin_port == addrp->sin_port || tsp->addr.sin_port == 0))
- break;
- }
-
- /* bump ref count if we found the server */
- if (tsp)
- cm_GetServerNoLock(tsp);
-
- /* drop big table lock */
- lock_ReleaseRead(&cm_serverLock);
-
- /* return what we found */
- return tsp;
+ return cm_FindServerByIP(addrp->sin_addr.s_addr, addrp->sin_port, type, locked);
}
cm_server_vols_t *cm_NewServerVols(void) {
return refCount;
}
+afs_uint32
+cm_ServerListSize(cm_serverRef_t* serversp)
+{
+ afs_uint32 count = 0;
+ cm_serverRef_t *tsrp;
+ lock_ObtainRead(&cm_serverLock);
+ for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+ count++;
+ }
+ lock_ReleaseRead(&cm_serverLock);
+ return count;
+}
LONG_PTR cm_ChecksumServerList(cm_serverRef_t *serversp)
{
void cm_InsertServerList(cm_serverRef_t** list, cm_serverRef_t* element)
{
cm_serverRef_t *current;
- unsigned short ipRank;
+ unsigned short rank;
lock_ObtainWrite(&cm_serverLock);
/*
cm_serverRef_t **currentp = list;
cm_serverRef_t **nextp = NULL;
cm_serverRef_t * next = NULL;
+ cm_server_t * serverp = NULL;
for (currentp = list; *currentp; currentp = nextp)
{
nextp = &(*currentp)->next;
+ /* obtain a refcnt on next in case cm_serverLock is dropped */
+ if (*nextp)
+ cm_GetServerRef(*nextp, TRUE);
if ((*currentp)->refCount == 0 &&
(*currentp)->status == srv_deleted) {
next = *nextp;
if ((*currentp)->volID)
cm_RemoveVolumeFromServer((*currentp)->server, (*currentp)->volID);
- cm_FreeServer((*currentp)->server);
+ serverp = (*currentp)->server;
free(*currentp);
nextp = &next;
+ /* cm_FreeServer will drop cm_serverLock if serverp->refCount == 0 */
+ cm_FreeServer(serverp);
}
+ /* drop the next refcnt obtained above. */
+ if (*nextp)
+ cm_PutServerRef(*nextp, TRUE);
}
}
goto done;
}
- ipRank = element->server->ipRank;
+ rank = element->server->activeRank;
/* insertion at the beginning of the list */
- if ((*list)->server->ipRank > ipRank)
+ if ((*list)->server->activeRank > rank)
{
element->next = *list;
*list = element;
/* find appropriate place to insert */
for ( current = *list; current->next; current = current->next)
{
- if ( current->next->server->ipRank > ipRank )
+ if ( current->next->server->activeRank > rank )
break;
}
element->next = current->next;
}
/* count the number of servers with the lowest rank */
- lowestRank = tsrp->server->ipRank;
+ lowestRank = tsrp->server->activeRank;
for ( count=1, tsrp=tsrp->next; tsrp; tsrp=tsrp->next)
{
- if ( tsrp->server->ipRank != lowestRank)
+ if ( tsrp->server->activeRank != lowestRank)
break;
else
count++;
*/
if (serverp->refCount == 0) {
if (!(serverp->flags & CM_SERVERFLAG_PREF_SET)) {
+ osi_QRemoveHT((osi_queue_t **)&cm_serversAllFirstp,
+ (osi_queue_t **)&cm_serversAllLastp,
+ &serverp->allq);
+
switch (serverp->type) {
case CM_SERVER_VLDB:
cm_numVldbServers--;
}
lock_FinalizeMutex(&serverp->mx);
- if ( cm_allServersp == serverp )
- cm_allServersp = serverp->allNextp;
- else {
- cm_server_t *tsp;
-
- for(tsp = cm_allServersp; tsp->allNextp; tsp=tsp->allNextp) {
- if ( tsp->allNextp == serverp ) {
- tsp->allNextp = serverp->allNextp;
- break;
- }
- }
- }
/* free the volid list */
for ( tsrvp = serverp->vols; tsrvp; tsrvp = nextp) {
}
}
+int cm_IsServerListEmpty(cm_serverRef_t *serversp)
+{
+ cm_serverRef_t *tsrp;
+ int allDeleted = 1;
+
+ if (serversp == NULL)
+ return CM_ERROR_EMPTY;
+
+ lock_ObtainRead(&cm_serverLock);
+ for (tsrp = serversp; tsrp; tsrp=tsrp->next) {
+ if (tsrp->status == srv_deleted)
+ continue;
+ allDeleted = 0;
+ break;
+ }
+ lock_ReleaseRead(&cm_serverLock);
+
+ return ( allDeleted ? CM_ERROR_EMPTY : 0 );
+}
+
void cm_FreeServerList(cm_serverRef_t** list, afs_uint32 flags)
{
cm_serverRef_t **current;
cm_serverRef_t **nextp;
cm_serverRef_t * next;
+ cm_server_t * serverp;
afs_int32 refCount;
lock_ObtainWrite(&cm_serverLock);
while (*current)
{
nextp = &(*current)->next;
+ /* obtain a refcnt on next in case cm_serverLock is dropped */
+ if (*nextp)
+ cm_GetServerRef(*nextp, TRUE);
refCount = cm_PutServerRef(*current, TRUE);
if (refCount == 0) {
next = *nextp;
if ((*current)->volID)
cm_RemoveVolumeFromServer((*current)->server, (*current)->volID);
- cm_FreeServer((*current)->server);
+ serverp = (*current)->server;
free(*current);
*current = next;
+ /* cm_FreeServer will drop cm_serverLock if serverp->refCount == 0 */
+ cm_FreeServer(serverp);
} else {
if (flags & CM_FREESERVERLIST_DELETE) {
(*current)->status = srv_deleted;
}
current = nextp;
}
+ /* drop the next refcnt obtained above. */
+ if (*current)
+ cm_PutServerRef(*current, TRUE);
}
done:
cookie, cm_numFileServers, cm_numVldbServers);
WriteFile(outputFile, output, (DWORD)strlen(output), &zilch, NULL);
- for (tsp = cm_allServersp; tsp; tsp=tsp->allNextp)
+ for (tsp = cm_serversAllFirstp;
+ tsp;
+ tsp = (cm_server_t *)osi_QNext(&tsp->allq))
{
char * type;
char * down;
"flags=0x%x waitCount=%u rank=%u downTime=\"%s\" refCount=%u\r\n",
cookie, tsp, tsp->cellp ? tsp->cellp->name : "", hoststr,
ntohs(tsp->addr.sin_port), uuidstr, type,
- tsp->capabilities, tsp->flags, tsp->waitCount, tsp->ipRank,
- (tsp->flags & CM_SERVERFLAG_DOWN) ? down : "up",
+ tsp->capabilities, tsp->flags, tsp->waitCount, tsp->activeRank,
+ (tsp->flags & CM_SERVERFLAG_DOWN) ? "down" : "up",
tsp->refCount);
WriteFile(outputFile, output, (DWORD)strlen(output), &zilch, NULL);
}
if (UuidEqual((UUID *)&srv1->uuid, (UUID *)&srv2->uuid, &status))
return 1;
} else {
- if (srv1->flags & CM_SERVERFLAG_UUID)
+ if (srv2->flags & CM_SERVERFLAG_UUID)
return 0;
/* Neither support UUID so perform an addr/port comparison */
git://git.openafs.org / openafs.git / blobdiff