Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

4101 lines
108 KiB

/*++
Copyright (c) 1996 Microsoft Corporation
Module Name:
cnpnet.c
Abstract:
Network management routines for the Cluster Network Protocol.
Author:
Mike Massa (mikemas) July 29, 1996
Revision History:
Who When What
-------- -------- ----------------------------------------------
mikemas 07-29-96 created
Notes:
--*/
#include "precomp.h"
#pragma hdrstop
#include "cnpnet.tmh"
#include <tdiinfo.h>
#include <tcpinfo.h>
#include <align.h>
#include <sspi.h>
//
// Global Data
//
LIST_ENTRY CnpNetworkList = {NULL, NULL};
LIST_ENTRY CnpDeletingNetworkList = {NULL, NULL};
#if DBG
CN_LOCK CnpNetworkListLock = {0,0};
#else // DBG
CN_LOCK CnpNetworkListLock = 0;
#endif // DBG
BOOLEAN CnpIsNetworkShutdownPending = FALSE;
PKEVENT CnpNetworkShutdownEvent = NULL;
USHORT CnpReservedClusnetPort = 0;
#ifdef ALLOC_PRAGMA
#pragma alloc_text(INIT, CnpLoadNetworks)
#pragma alloc_text(PAGE, CnpInitializeNetworks)
#endif // ALLOC_PRAGMA
//
// Private utiltity routines
//
#define CnpIpAddrPrintArgs(_ip) \
((_ip >> 0 ) & 0xff), \
((_ip >> 8 ) & 0xff), \
((_ip >> 16) & 0xff), \
((_ip >> 24) & 0xff)
#define CnpIsInternalMulticastNetwork(_network) \
(((_network)->State == ClusnetNetworkStateOnline) && \
(!CnpIsNetworkRestricted((_network))) && \
(CnpIsNetworkMulticastCapable((_network))))
VOID
CnpMulticastGetReachableNodesLocked(
OUT CX_CLUSTERSCREEN * McastReachableNodes,
OUT ULONG * McastReachableCount
)
{
PLIST_ENTRY entry;
PCNP_NETWORK network = NULL;
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
if (!IsListEmpty(&CnpNetworkList)) {
entry = CnpNetworkList.Flink;
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
//
// The old screen and count are only valid if
// this is a valid internal network.
//
if (CnpIsInternalMulticastNetwork(network)) {
*McastReachableNodes = network->McastReachableNodes;
*McastReachableCount = network->McastReachableCount;
} else {
network = NULL;
}
}
if (network == NULL) {
RtlZeroMemory(McastReachableNodes, sizeof(*McastReachableNodes));
*McastReachableCount = 0;
}
return;
} // CnpMulticastGetReachableNodesLocked
BOOLEAN
CnpRemoveNetworkListEntryLocked(
IN PCNP_NETWORK Network,
IN BOOLEAN RaiseEvent,
OUT CX_CLUSTERSCREEN * McastReachableNodes OPTIONAL
)
/*++
Routine Description:
Remove Network from the network list and return the new
multicast reachable mask.
Return value:
TRUE if the reachable set changed
Notes:
Called and returns with network list lock held.
--*/
{
ULONG count;
BOOLEAN setChanged;
CX_CLUSTERSCREEN oldScreen;
CX_CLUSTERSCREEN newScreen;
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
CnpMulticastGetReachableNodesLocked(&oldScreen, &count);
RemoveEntryList(&(Network->Linkage));
Network->Flags &= ~CNP_NET_FLAG_MCASTSORTED;
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
setChanged = (BOOLEAN)
(oldScreen.UlongScreen != newScreen.UlongScreen);
if (RaiseEvent && setChanged) {
CnTrace(CNP_NET_DETAIL, CnpTraceMulticastReachEventRemove,
"[CNP] Issuing event for new multicast "
"reachable set (%lx) after removing "
"network %u.",
newScreen.UlongScreen,
Network->Id
);
CnIssueEvent(
ClusnetEventMulticastSet,
newScreen.UlongScreen,
0
);
}
if (McastReachableNodes != NULL) {
*McastReachableNodes = newScreen;
}
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
return(setChanged);
} // CnpRemoveNetworkListEntryLocked
BOOLEAN
CnpIsBetterMulticastNetwork(
IN PCNP_NETWORK Network1,
IN PCNP_NETWORK Network2
)
/*++
Routine Description:
Compares two networks according to multicast reachability
criteria:
1. online/registered AND
not restricted (e.g. enabled for intracluster comm) AND
not disconnected AND
multicast-enabled
2. priority
3. number of multicast reachable nodes
Return value:
TRUE if Network1 is better than Network2
--*/
{
if (!CnpIsInternalMulticastNetwork(Network1)) {
return(FALSE);
}
if (!CnpIsInternalMulticastNetwork(Network2)) {
return(TRUE);
}
//
// Both networks are equal with respect to basic
// multicast requirements.
//
// Now compare the priority.
//
if (CnpIsEqualPriority(Network1->Priority, Network2->Priority)) {
//
// The priority is the same. Although this is unexpected,
// we now compare the number of nodes reachable by
// multicast.
//
return(Network1->McastReachableCount > Network2->McastReachableCount);
} else {
return(CnpIsHigherPriority(Network1->Priority, Network2->Priority));
}
} // CnpIsBetterMulticastNetwork
BOOLEAN
CnpSortMulticastNetworkLocked(
IN PCNP_NETWORK Network,
IN BOOLEAN RaiseEvent,
OUT CX_CLUSTERSCREEN * NewMcastReachableNodes OPTIONAL
)
/*++
Routine Description:
Positions Network in network list according to multicast
reachability. Network must already be inserted in the
network list.
The network list is always sorted, but it is possible
for one network in the list to be "perturbed". In this
case, that entry must be repositioned correctly. This
routine handles repositioning.
Returns new screen through NewMcastReachableNodes.
Return value:
TRUE if number of reachable nodes changes.
Notes:
Called and returns with network list locked.
--*/
{
ULONG count;
CX_CLUSTERSCREEN oldScreen;
CX_CLUSTERSCREEN newScreen;
BOOLEAN newScreenValid = FALSE;
PLIST_ENTRY entry;
PCNP_NETWORK network = NULL;
KIRQL irql;
BOOLEAN move = FALSE;
BOOLEAN setChanged = FALSE;
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
//
// If the network has already been removed from the
// sorted list, there is no sense in resorting it.
//
if (CnpIsNetworkMulticastSorted(Network)) {
//
// Remember the current screen and count to detect
// changes.
//
CnpMulticastGetReachableNodesLocked(&oldScreen, &count);
//
// Check if it needs to be moved up.
//
for (entry = Network->Linkage.Blink;
entry != &CnpNetworkList;
entry = entry->Blink) {
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
if (CnpIsBetterMulticastNetwork(Network, network)) {
move = TRUE;
} else {
break;
}
}
if (move) {
RemoveEntryList(&(Network->Linkage));
InsertHeadList(entry, &(Network->Linkage));
} else {
//
// Check if it needs to be moved down.
//
for (entry = Network->Linkage.Flink;
entry != &CnpNetworkList;
entry = entry->Flink) {
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
if (CnpIsBetterMulticastNetwork(network, Network)) {
move = TRUE;
} else {
break;
}
}
if (move) {
RemoveEntryList(&(Network->Linkage));
InsertTailList(entry, &(Network->Linkage));
}
}
//
// Determine if the set of reachable nodes has changed.
//
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
newScreenValid = TRUE;
setChanged = (BOOLEAN)
(oldScreen.UlongScreen != newScreen.UlongScreen);
}
if (RaiseEvent && setChanged) {
if (!newScreenValid) {
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
newScreenValid = TRUE;
}
CnTrace(CNP_NET_DETAIL, CnpTraceMulticastReachEventSort,
"[CNP] Issuing event for new multicast "
"reachable set (%lx) after sorting "
"network %u.",
newScreen.UlongScreen,
Network->Id
);
CnIssueEvent(
ClusnetEventMulticastSet,
newScreen.UlongScreen,
0
);
}
if (NewMcastReachableNodes != NULL) {
if (!newScreenValid) {
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
newScreenValid = TRUE;
}
*NewMcastReachableNodes = newScreen;
}
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
return(setChanged);
} // CnpSortMulticastNetworkLocked
BOOLEAN
CnpMulticastChangeNodeReachabilityLocked(
IN PCNP_NETWORK Network,
IN PCNP_NODE Node,
IN BOOLEAN Reachable,
IN BOOLEAN RaiseEvent,
OUT CX_CLUSTERSCREEN * NewMcastReachableNodes OPTIONAL
)
/*++
Routine Description:
Changes the multicast reachability state of Node
on Network.
If the set of reachable nodes changes, returns
the new screen through NewMcastReachableNodes.
Return value:
TRUE if set of reachable nodes changes.
Notes:
Called and returns with node lock held.
Called and returns with network list lock held.
--*/
{
KIRQL irql;
BOOLEAN netSetChanged = FALSE;
BOOLEAN setChanged = FALSE;
CX_CLUSTERSCREEN oldScreen;
CX_CLUSTERSCREEN newScreen;
BOOLEAN newScreenValid = FALSE;
ULONG count;
CnVerifyCpuLockMask(
CNP_NODE_OBJECT_LOCK | CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
if (Reachable) {
if (Node != CnpLocalNode) {
if (!CnpClusterScreenMember(
Network->McastReachableNodes.ClusterScreen,
INT_NODE(Node->Id)
)) {
//
// Remember the current screen and count to detect
// changes.
//
CnpMulticastGetReachableNodesLocked(&oldScreen, &count);
CnpClusterScreenInsert(
Network->McastReachableNodes.ClusterScreen,
INT_NODE(Node->Id)
);
Network->McastReachableCount++;
netSetChanged = TRUE;
}
}
} else {
if (Node == CnpLocalNode) {
//
// Remember the current screen and count to detect
// changes.
//
CnpMulticastGetReachableNodesLocked(&oldScreen, &count);
//
// The local interface on this network
// no longer speaks multicast. Declare all
// other nodes unreachable.
//
CnpNetworkResetMcastReachableNodes(Network);
if (Network->McastReachableCount != 0) {
netSetChanged = TRUE;
}
Network->McastReachableCount = 0;
} else {
if (CnpClusterScreenMember(
Network->McastReachableNodes.ClusterScreen,
INT_NODE(Node->Id)
)) {
//
// Remember the current screen and count to detect
// changes.
//
CnpMulticastGetReachableNodesLocked(&oldScreen, &count);
CnpClusterScreenDelete(
Network->McastReachableNodes.ClusterScreen,
INT_NODE(Node->Id)
);
Network->McastReachableCount--;
netSetChanged = TRUE;
}
}
}
if (netSetChanged) {
CnpSortMulticastNetworkLocked(Network, FALSE, &newScreen);
newScreenValid = TRUE;
setChanged = (BOOLEAN)(oldScreen.UlongScreen != newScreen.UlongScreen);
}
if (RaiseEvent && setChanged) {
if (!newScreenValid) {
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
newScreenValid = TRUE;
}
CnTrace(CNP_NET_DETAIL, CnpTraceMulticastReachEventReach,
"[CNP] Issuing event for new multicast "
"reachable set (%lx) after setting "
"reachability for network %u to %!bool!.",
newScreen.UlongScreen,
Network->Id, Reachable
);
CnIssueEvent(
ClusnetEventMulticastSet,
newScreen.UlongScreen,
0
);
}
if (NewMcastReachableNodes != NULL) {
if (!newScreenValid) {
CnpMulticastGetReachableNodesLocked(&newScreen, &count);
newScreenValid = TRUE;
}
*NewMcastReachableNodes = newScreen;
}
CnVerifyCpuLockMask(
CNP_NODE_OBJECT_LOCK | CNP_NETWORK_LIST_LOCK, // required
0, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
return(setChanged);
} // CnpMulticastChangeNodeReachabilityLocked
PCNP_NETWORK
CnpLockedFindNetwork(
IN CL_NETWORK_ID NetworkId,
IN CN_IRQL ListIrql
)
/*++
Routine Description:
Searches the network list for a specified network object.
Arguments:
NetworkId - The ID of the network object to locate.
ListIrql - The IRQL level at which the network list lock was
acquired before calling this routine.
Return Value:
A pointer to the requested network object, if it exists.
NULL otherwise.
Notes:
Called with CnpNetworkListLock held.
Returns with CnpNetworkListLock released.
If return value is non-NULL, returns with network object lock held.
--*/
{
PLIST_ENTRY entry;
CN_IRQL networkIrql;
PCNP_NETWORK network = NULL;
CnVerifyCpuLockMask(
CNP_NETWORK_LIST_LOCK, // Required
0, // Forbidden
CNP_NETWORK_LIST_LOCK_MAX // Maximum
);
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLock(&(network->Lock), &networkIrql);
if (NetworkId == network->Id) {
CnReleaseLock(&CnpNetworkListLock, networkIrql);
network->Irql = ListIrql;
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
CNP_NETWORK_LIST_LOCK, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
return(network);
}
CnReleaseLock(&(network->Lock), networkIrql);
}
CnReleaseLock(&CnpNetworkListLock, ListIrql);
CnVerifyCpuLockMask(
0, // Required
(CNP_NETWORK_LIST_LOCK | CNP_NETWORK_OBJECT_LOCK), // Forbidden
CNP_NODE_OBJECT_LOCK_MAX // Maximum
);
return(NULL);
} // CnpLockedFindNetwork
VOID
CnpOfflineNetwork(
PCNP_NETWORK Network
)
/*++
Notes:
Called with network object lock held.
Returns with network object lock released.
May not be called while holding any higher-ranked locks.
--*/
{
NTSTATUS status;
PFILE_OBJECT fileObject = NULL;
PDEVICE_OBJECT deviceObject = NULL;
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
(ULONG) ~(CNP_NETWORK_OBJECT_LOCK), // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
CnAssert(Network->State >= ClusnetNetworkStateOnlinePending);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Offline of network %u pending....\n",
Network->Id
));
}
Network->State = ClusnetNetworkStateOfflinePending;
//
// Get the UDP file and device objects associated
// with this network to clear the receive event
// handler. Referencing the file object will keep
// both the file and device objects from going away,
// since the I/O mgr maintains a ref on the device
// object for the file object.
//
fileObject = Network->DatagramFileObject;
if (fileObject != NULL) {
ObReferenceObject(fileObject);
}
deviceObject = Network->DatagramDeviceObject;
CnReleaseLock(&(Network->Lock), Network->Irql);
CnTrace(
CNP_NET_DETAIL, CnpTraceNetworkOfflinePending,
"[CNP] Offline of network %u pending.",
Network->Id
);
//
// If the network is still on the sorted network list,
// re-sort.
//
CnpSortMulticastNetwork(Network, TRUE, NULL);
//
// Take all of the interfaces on this network offline.
//
// Note that the network cannot go away while we do this
// because we still hold an active reference on it.
//
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Taking all interfaces on network %u offline...\n",
Network->Id
));
}
CnpWalkInterfacesOnNetwork(Network, CnpOfflineInterfaceWrapper);
//
// Stop the incoming data flow by clearing the receive
// datagram TDI event handler.
//
if (fileObject != NULL && deviceObject != NULL) {
status = CnpTdiSetEventHandler(
fileObject,
deviceObject,
TDI_EVENT_RECEIVE_DATAGRAM,
NULL, // handler
NULL, // context
NULL // IRP for re-use
);
if (!NT_SUCCESS(status)) {
CnTrace(CNP_NET_DETAIL, CnpTraceNetworkClearRecv,
"[CNP] Failed to clear receive datagram handler "
"for network ID %u, status %!status!",
Network->Id, // LOGULONG
status // LOGSTATUS
);
// Non-fatal error. All will be cleaned up when we
// close the file object.
}
}
//
// Drop the reference we took on the file object.
//
if (fileObject != NULL) {
ObDereferenceObject(fileObject);
}
CnAcquireLock(&(Network->Lock), &(Network->Irql));
//
// Remove the initial active reference. When the active
// reference count goes to zero, the network will be taken
// offline and the irp completed.
//
// The network object lock will be released by
// the dereference.
//
CnpActiveDereferenceNetwork(Network);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return;
} // CnpOfflineNetwork
VOID
CnpOfflineNetworkWorkRoutine(
IN PVOID Parameter
)
/*++
Routine Description:
Performs the actual work involved in taking a network offline.
This routine runs in the context of an ExWorkerThread.
Arguments:
Parameter - A pointer to the network object on which to operate.
Return Value:
None.
--*/
{
NTSTATUS status;
HANDLE handle = NULL;
PFILE_OBJECT fileObject = NULL;
PIRP offlineIrp;
PCNP_NETWORK network = Parameter;
CnAssert(KeGetCurrentIrql() == PASSIVE_LEVEL);
CnAssert(network->State == ClusnetNetworkStateOfflinePending);
CnAssert(CnSystemProcess == (PKPROCESS) IoGetCurrentProcess());
CnAcquireLock(&(network->Lock), &(network->Irql));
handle = network->DatagramHandle;
network->DatagramHandle = NULL;
fileObject = network->DatagramFileObject;
network->DatagramFileObject = NULL;
network->DatagramDeviceObject = NULL;
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Taking network %u offline...\n", network->Id));
}
CnReleaseLock(&(network->Lock), network->Irql);
CnTrace(CNP_NET_DETAIL, CnpTraceNetworkTakingOffline,
"[CNP] Taking network %u offline, dgram handle %p, "
"dgram fileobj %p.",
network->Id, // LOGULONG
handle, // LOGHANDLE
fileObject // LOGPTR
);
if (fileObject != NULL) {
ObDereferenceObject(fileObject);
}
if (handle != NULL) {
status = ZwClose(handle);
IF_CNDBG(CN_DEBUG_CONFIG) {
if (!NT_SUCCESS(status)) {
CNPRINT(("[CNP] Failed to close handle for network %u, "
"status %lx.\n",
network->Id, status));
}
}
CnAssert(NT_SUCCESS(status));
CnTrace(CNP_NET_DETAIL, CnpTraceNetworkClosed,
"[CNP] Closed handle %p for network ID %u, status %!status!",
handle, // LOGHANDLE
network->Id, // LOGULONG
status // LOGSTATUS
);
}
CnAcquireLock(&(network->Lock), &(network->Irql));
CnAssert(network->State == ClusnetNetworkStateOfflinePending);
network->State = ClusnetNetworkStateOffline;
offlineIrp = network->PendingOfflineIrp;
network->PendingOfflineIrp = NULL;
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Network %u is now offline.\n", network->Id));
}
//
// Remove the active reference from the base refcount.
// This releases the network object lock.
//
CnpDereferenceNetwork(network);
if (offlineIrp != NULL) {
CN_IRQL cancelIrql;
CnAcquireCancelSpinLock(&cancelIrql);
offlineIrp->CancelIrql = cancelIrql;
CnCompletePendingRequest(offlineIrp, STATUS_SUCCESS, 0);
}
CnAssert(KeGetCurrentIrql() == PASSIVE_LEVEL);
return;
} // CnpOfflineNetworkWorkRoutine
VOID
CnpDeleteNetwork(
PCNP_NETWORK Network,
CN_IRQL NetworkListIrql
)
/*++
Notes:
Called with the CnpNetworkListLock and network object lock held.
Returns with both locks released.
--*/
{
NTSTATUS status;
ULONG i;
PCNP_INTERFACE interface;
CL_NETWORK_ID networkId = Network->Id;
CnVerifyCpuLockMask(
(CNP_NETWORK_LIST_LOCK | CNP_NETWORK_OBJECT_LOCK), // Required
0, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Deleting network %u\n", Network->Id));
}
//
// Move the network to the deleting list. Once we do this,
// no new threads can reference the network object.
//
CnpRemoveNetworkListEntryLocked(Network, TRUE, NULL);
InsertTailList(&CnpDeletingNetworkList, &(Network->Linkage));
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Moved network %u to deleting list\n",
Network->Id
));
}
CnReleaseLockFromDpc(&CnpNetworkListLock);
Network->Irql = NetworkListIrql;
Network->Flags |= CNP_NET_FLAG_DELETING;
if (Network->State >= ClusnetNetworkStateOnlinePending) {
//
// Take the network offline. This will force all of the
// associated interfaces offline as well.
//
// This will release the network object lock.
//
CnpOfflineNetwork(Network);
}
else {
CnReleaseLock(&(Network->Lock), Network->Irql);
}
//
// Delete all the interfaces on this network.
//
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Deleting all interfaces on network %u...\n",
Network->Id
));
}
CnpWalkInterfacesOnNetwork(Network, CnpDeleteInterface);
//
// Remove the initial reference on the object. The object will be
// destroyed when the reference count goes to zero. The delete irp
// will be completed at that time.
//
CnAcquireLock(&(Network->Lock), &(Network->Irql));
CnpDereferenceNetwork(Network);
CnVerifyCpuLockMask(
0, // Required
(CNP_NETWORK_OBJECT_LOCK | CNP_NETWORK_OBJECT_LOCK), // Forbidden
CNP_NODE_OBJECT_LOCK_MAX // Maximum
);
return;
} // CnpDeleteNework
VOID
CnpDestroyNetworkWorkRoutine(
IN PVOID Parameter
)
/*++
Routine Description:
Performs the actual work involved in destroying a network.
This routine runs in the context of an ExWorkerThread.
Arguments:
Parameter - A pointer to the network object on which to operate.
Return Value:
None.
--*/
{
PLIST_ENTRY entry;
CN_IRQL listIrql;
BOOLEAN setCleanupEvent = FALSE;
PCNP_NETWORK network = Parameter;
CnAssert(KeGetCurrentIrql() == PASSIVE_LEVEL);
CnAssert(network->State == ClusnetNetworkStateOffline);
CnAssert(CnSystemProcess == (PKPROCESS) IoGetCurrentProcess());
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Destroying network %u\n", network->Id));
}
CnAcquireLock(&CnpNetworkListLock, &listIrql);
#if DBG
{
PCNP_NETWORK oldNetwork = NULL;
//
// Verify that the network object is on the deleting list.
//
for (entry = CnpDeletingNetworkList.Flink;
entry != &CnpDeletingNetworkList;
entry = entry->Flink
)
{
oldNetwork = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
if (oldNetwork == network) {
break;
}
}
CnAssert(oldNetwork == network);
}
#endif // DBG
RemoveEntryList(&(network->Linkage));
if (CnpIsNetworkShutdownPending) {
if (IsListEmpty(&CnpDeletingNetworkList)) {
setCleanupEvent = TRUE;
}
}
CnReleaseLock(&CnpNetworkListLock, listIrql);
if (network->PendingDeleteIrp != NULL) {
CnAcquireCancelSpinLock(&(network->PendingDeleteIrp->CancelIrql));
CnCompletePendingRequest(
network->PendingDeleteIrp,
STATUS_SUCCESS,
0
);
//
// The IoCancelSpinLock was released by CnCompletePendingRequest()
//
}
if (network->CurrentMcastGroup != NULL) {
CnpDereferenceMulticastGroup(network->CurrentMcastGroup);
network->CurrentMcastGroup = NULL;
}
if (network->PreviousMcastGroup != NULL) {
CnpDereferenceMulticastGroup(network->PreviousMcastGroup);
network->PreviousMcastGroup = NULL;
}
CnFreePool(network);
if (setCleanupEvent) {
IF_CNDBG(CN_DEBUG_INIT) {
CNPRINT(("[CNP] Setting network cleanup event.\n"));
}
KeSetEvent(CnpNetworkShutdownEvent, 0, FALSE);
}
CnAssert(KeGetCurrentIrql() == PASSIVE_LEVEL);
return;
} // CnpDestroyNetworkWorkRoutine
//
// Routines exported within CNP
//
PCNP_NETWORK
CnpFindNetwork(
IN CL_NETWORK_ID NetworkId
)
/*++
Notes:
--*/
{
CN_IRQL listIrql;
CnAcquireLock(&CnpNetworkListLock, &listIrql);
return(CnpLockedFindNetwork(NetworkId, listIrql));
} // CnpFindNetwork
VOID
CnpReferenceNetwork(
PCNP_NETWORK Network
)
/*++
Notes:
Called with network object lock held.
--*/
{
CnAssert(Network->RefCount != 0xFFFFFFFF);
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
0, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
Network->RefCount++;
IF_CNDBG(CN_DEBUG_CNPREF) {
CNPRINT((
"[CNP] Referencing network %u, new refcount %u\n",
Network->Id,
Network->RefCount
));
}
return;
} // CnpReferenceNetwork
VOID
CnpDereferenceNetwork(
PCNP_NETWORK Network
)
/*++
Notes:
Called with network object lock held.
Returns with network object lock released.
Sometimes called with a node object lock held as well.
--*/
{
PLIST_ENTRY entry;
CN_IRQL listIrql;
BOOLEAN setCleanupEvent = FALSE;
ULONG newRefCount;
PCNP_NETWORK network;
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
0, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
CnAssert(Network->RefCount != 0);
newRefCount = --(Network->RefCount);
IF_CNDBG(CN_DEBUG_CNPREF) {
CNPRINT((
"[CNP] Dereferencing network %u, new refcount %u\n",
Network->Id,
Network->RefCount
));
}
CnReleaseLock(&(Network->Lock), Network->Irql);
if (newRefCount > 0) {
CnVerifyCpuLockMask(
0, // Required
CNP_NETWORK_OBJECT_LOCK, // Forbidden
CNP_NETWORK_LIST_LOCK_MAX // Maximum
);
return;
}
CnAssert(Network->ActiveRefCount == 0);
CnAssert(Network->State == ClusnetNetworkStateOffline);
CnAssert(Network->DatagramHandle == NULL);
//
// Schedule an ExWorkerThread to destroy the network.
// We do this because we don't know if a higher-level lock,
// such as a node object lock, is held when this routine is
// called. We may need to acquire the IoCancelSpinLock,
// which must be acquired before a node lock, in
// order to complete a deregister Irp.
//
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Posting destroy work item for network %u.\n",
Network->Id
));
}
ExInitializeWorkItem(
&(Network->ExWorkItem),
CnpDestroyNetworkWorkRoutine,
Network
);
ExQueueWorkItem(&(Network->ExWorkItem), DelayedWorkQueue);
CnVerifyCpuLockMask(
0, // Required
CNP_NETWORK_OBJECT_LOCK, // Forbidden
CNP_NETWORK_LIST_LOCK_MAX // Maximum
);
return;
} // CnpDereferenceNetwork
ULONG
CnpActiveReferenceNetwork(
PCNP_NETWORK Network
)
/*++
Return value:
New ref count.
Notes:
Called with network object lock held.
Will not allow reference if current refcount is zero.
--*/
{
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
0, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
CnAssert(Network->ActiveRefCount != 0xFFFFFFFF);
CnAssert(Network->RefCount != 0);
if (Network->ActiveRefCount == 0) {
CnTrace(CNP_NET_DETAIL, CnpActiveRefNetFailed,
"[CNP] Cannot take active reference on "
"network %u because the active refcount "
"is already zero.",
Network->Id
);
return(Network->ActiveRefCount);
}
return (++Network->ActiveRefCount);
} // CnpActiveReferenceNetwork
VOID
CnpActiveDereferenceNetwork(
PCNP_NETWORK Network
)
/*++
Notes:
Called with network object lock held.
Returns with network object lock released.
--*/
{
ULONG newRefCount;
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // Required
0, // Forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // Maximum
);
CnAssert(Network->ActiveRefCount != 0);
CnAssert(Network->State != ClusnetNetworkStateOffline);
newRefCount = --(Network->ActiveRefCount);
CnReleaseLock(&(Network->Lock), Network->Irql);
if (newRefCount > 0) {
CnVerifyCpuLockMask(
0, // Required
CNP_NETWORK_OBJECT_LOCK, // Forbidden
CNP_NETWORK_LIST_LOCK_MAX // Maximum
);
return;
}
//
// The network's active reference count has gone to zero.
//
CnAssert(Network->State == ClusnetNetworkStateOfflinePending);
//
// Schedule an ExWorkerThread to take the network offline.
// We do this because we don't know if a higher-level lock,
// such as a node object lock, is held when this routine is
// called. The base transport file handle must be closed at
// PASSIVE_LEVEL. We may also need to acquire the IoCancelSpinLock
// in order to complete an Offline Irp.
//
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Posting offline work item for network %u.\n",
Network->Id
));
}
CnTrace(
CNP_NET_DETAIL, CnpTraceNetworkSchedulingOffline,
"[CNP] Scheduling offline of network %u.",
Network->Id
);
ExInitializeWorkItem(
&(Network->ExWorkItem),
CnpOfflineNetworkWorkRoutine,
Network
);
ExQueueWorkItem(&(Network->ExWorkItem), DelayedWorkQueue);
CnVerifyCpuLockMask(
0, // Required
CNP_NETWORK_OBJECT_LOCK, // Forbidden
CNP_NETWORK_LIST_LOCK_MAX // Maximum
);
return;
} // CnpActiveDereferenceNetwork
NTSTATUS
CnpAllocateMulticastGroup(
IN ULONG Brand,
IN PTRANSPORT_ADDRESS TdiMulticastAddress,
IN ULONG TdiMulticastAddressLength,
IN PVOID Key,
IN ULONG KeyLength,
OUT PCNP_MULTICAST_GROUP * Group
)
/*++
Routine Description:
Allocates and initializes a network multicast group
structure.
--*/
{
PCNP_MULTICAST_GROUP group;
ULONG groupSize;
UCHAR keyBuffer[DES_BLOCKLEN];
PUCHAR key;
//
// Allocate the data structure.
//
groupSize = sizeof(CNP_MULTICAST_GROUP);
if (TdiMulticastAddressLength != 0) {
groupSize = ROUND_UP_COUNT(groupSize,
TYPE_ALIGNMENT(TRANSPORT_ADDRESS)) +
TdiMulticastAddressLength;
}
if (KeyLength != 0) {
groupSize = ROUND_UP_COUNT(groupSize, TYPE_ALIGNMENT(PVOID)) +
KeyLength;
}
group = CnAllocatePool(groupSize);
if (group == NULL) {
return(STATUS_INSUFFICIENT_RESOURCES);
}
//
// Fill in parameter fields.
//
group->McastNetworkBrand = Brand;
group->McastTdiAddress = (PTRANSPORT_ADDRESS)
ROUND_UP_POINTER((PUCHAR)group + sizeof(CNP_MULTICAST_GROUP),
TYPE_ALIGNMENT(TRANSPORT_ADDRESS));
group->McastTdiAddressLength = TdiMulticastAddressLength;
RtlCopyMemory(
group->McastTdiAddress,
TdiMulticastAddress,
TdiMulticastAddressLength
);
group->Key = (PVOID)
ROUND_UP_POINTER((PUCHAR)group->McastTdiAddress
+ TdiMulticastAddressLength,
TYPE_ALIGNMENT(PVOID));
group->KeyLength = KeyLength;
RtlCopyMemory(
group->Key,
Key,
KeyLength
);
group->SignatureLength = CX_SIGNATURE_LENGTH;
//
// Set the initial refcount to 1.
//
group->RefCount = 1;
*Group = group;
return(STATUS_SUCCESS);
} // CnpAllocateMulticastGroup
VOID
CnpFreeMulticastGroup(
IN PCNP_MULTICAST_GROUP Group
)
{
if (Group != NULL) {
CnFreePool(Group);
}
return;
} // CnpFreeMulticastGroup
NTSTATUS
CnpConfigureBasicMulticastSettings(
IN HANDLE Handle,
IN PFILE_OBJECT FileObject,
IN PDEVICE_OBJECT DeviceObject,
IN PTDI_ADDRESS_INFO TdiBindAddressInfo,
IN ULONG McastTtl,
IN UCHAR McastLoop,
IN PIRP Irp
)
/*++
Routine Description:
Set basic multicast parameters on the address object
represented by Handle, FileObject, and DeviceObject
using the interface represented by TdiBindAddressInfo.
Notes:
This routine attaches to the system process when using
handles, so it should not be called pre-attached.
--*/
{
UDPMCastIFReq mcastIfReq;
ULONG ifBindIp;
BOOLEAN attached = FALSE;
NTSTATUS status;
//
// Set this interface for outgoing multicast traffic.
//
ifBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)&(TdiBindAddressInfo->Address))
->Address[0].Address[0].in_addr)
)
);
mcastIfReq.umi_addr = ifBindIp;
KeAttachProcess(CnSystemProcess);
attached = TRUE;
status = CnpSetTcpInfoEx(
Handle,
CL_TL_ENTITY,
INFO_CLASS_PROTOCOL,
INFO_TYPE_ADDRESS_OBJECT,
AO_OPTION_MCASTIF,
&mcastIfReq,
sizeof(mcastIfReq)
);
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] Set mcast interface for "
"AO handle %p, IF %d.%d.%d.%d, status %x.\n",
Handle,
CnpIpAddrPrintArgs(ifBindIp),
status
));
}
if (!NT_SUCCESS(status)) {
goto error_exit;
}
status = CnpSetTcpInfoEx(
Handle,
CL_TL_ENTITY,
INFO_CLASS_PROTOCOL,
INFO_TYPE_ADDRESS_OBJECT,
AO_OPTION_MCASTTTL,
&McastTtl,
sizeof(McastTtl)
);
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] Set mcast TTL to %d on "
"AO handle %p, IF %d.%d.%d.%d, "
"status %x.\n",
McastTtl,
Handle,
CnpIpAddrPrintArgs(ifBindIp),
status
));
}
if (!NT_SUCCESS(status)) {
goto error_exit;
}
status = CnpSetTcpInfoEx(
Handle,
CL_TL_ENTITY,
INFO_CLASS_PROTOCOL,
INFO_TYPE_ADDRESS_OBJECT,
AO_OPTION_MCASTLOOP,
&McastLoop,
sizeof(McastLoop)
);
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] Set mcast loopback flag to %d on "
"AO handle %p, IF %d.%d.%d.%d, status %x.\n",
McastLoop,
Handle,
CnpIpAddrPrintArgs(ifBindIp),
status
));
}
if (!NT_SUCCESS(status)) {
goto error_exit;
}
error_exit:
if (attached) {
KeDetachProcess();
attached = FALSE;
}
return(status);
} // CnpConfigureBasicMulticastSettings
NTSTATUS
CnpAddRemoveMulticastAddress(
IN HANDLE Handle,
IN PFILE_OBJECT FileObject,
IN PDEVICE_OBJECT DeviceObject,
IN PTDI_ADDRESS_INFO TdiBindAddressInfo,
IN PTRANSPORT_ADDRESS TdiMcastBindAddress,
IN ULONG OpId,
IN PIRP Irp
)
/*++
Routine Description:
Add or remove the multicast address specified by
TdiMcastBindAddress from the interface specified
by TdiBindAddressInfo.
Arguments:
OpId - either AO_OPTION_ADD_MCAST or AO_OPTION_DEL_MCAST
Notes:
This routine attaches to the system process when using
handles, so it should not be called pre-attached.
--*/
{
UDPMCastReq mcastAddDelReq;
ULONG mcastBindIp;
ULONG ifBindIp;
BOOLEAN attached = FALSE;
NTSTATUS status;
mcastBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)TdiMcastBindAddress)
->Address[0].Address[0].in_addr)
)
);
ifBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)&(TdiBindAddressInfo->Address))
->Address[0].Address[0].in_addr)
)
);
mcastAddDelReq.umr_addr = mcastBindIp;
mcastAddDelReq.umr_if = ifBindIp;
KeAttachProcess(CnSystemProcess);
attached = TRUE;
status = CnpSetTcpInfoEx(
Handle,
CL_TL_ENTITY,
INFO_CLASS_PROTOCOL,
INFO_TYPE_ADDRESS_OBJECT,
OpId,
&mcastAddDelReq,
sizeof(mcastAddDelReq)
);
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] Adjusted mcast binding on "
"interface for AO handle %p, "
"IF %d.%d.%d.%d, mcast addr %d.%d.%d.%d, "
"OpId %d, status %x.\n",
Handle,
CnpIpAddrPrintArgs(ifBindIp),
CnpIpAddrPrintArgs(mcastBindIp),
OpId,
status
));
}
if (!NT_SUCCESS(status)) {
goto error_exit;
}
error_exit:
if (attached) {
KeDetachProcess();
attached = FALSE;
}
return(status);
} // CnpAddRemoveMulticastAddress
VOID
CnpStartInterfaceMcastTransition(
PCNP_INTERFACE Interface
)
/*++
Routine Description:
Called during a multicast group transition. Clears
the multicast received flag and enables discovery.
Arguments:
Interface - A pointer to the interface to change.
Return Value:
None.
Notes:
Conforms to the calling convention for
PCNP_INTERFACE_UPDATE_ROUTINE.
Called with associated node and network locks held.
Returns with network lock released.
--*/
{
if (Interface->Node != CnpLocalNode) {
CnpInterfaceClearReceivedMulticast(Interface);
Interface->McastDiscoverCount = CNP_INTERFACE_MCAST_DISCOVERY;
}
CnReleaseLock(&(Interface->Network->Lock), Interface->Network->Irql);
return;
} // CnpStartInterfaceMcastTransition
//
// Cluster Transport Public Routines
//
NTSTATUS
CnpLoadNetworks(
VOID
)
/*++
Routine Description:
Called when the Cluster Network driver is loading. Initializes
static network-related data structures.
Arguments:
None.
Return Value:
None.
--*/
{
InitializeListHead(&CnpNetworkList);
InitializeListHead(&CnpDeletingNetworkList);
CnInitializeLock(&CnpNetworkListLock, CNP_NETWORK_LIST_LOCK);
return(STATUS_SUCCESS);
} // CnpLoadNetworks
NTSTATUS
CnpInitializeNetworks(
VOID
)
/*++
Routine Description:
Called when the Cluster Network driver is being (re)initialized.
Initializes dynamic network-related data structures.
Arguments:
None.
Return Value:
None.
--*/
{
PAGED_CODE();
CnAssert(CnpNetworkShutdownEvent == NULL);
CnAssert(IsListEmpty(&CnpNetworkList));
CnAssert(IsListEmpty(&CnpDeletingNetworkList));
CnpNetworkShutdownEvent = CnAllocatePool(sizeof(KEVENT));
if (CnpNetworkShutdownEvent == NULL) {
return(STATUS_INSUFFICIENT_RESOURCES);
}
KeInitializeEvent(CnpNetworkShutdownEvent, NotificationEvent, FALSE);
CnpIsNetworkShutdownPending = FALSE;
return(STATUS_SUCCESS);
} // CnpInitializeNetworks
VOID
CnpShutdownNetworks(
VOID
)
/*++
Routine Description:
Called when a shutdown request is issued to the Cluster Network
Driver. Deletes all network objects.
Arguments:
None.
Return Value:
None.
--*/
{
PLIST_ENTRY entry;
CN_IRQL listIrql;
CN_IRQL networkIrql;
PCNP_NETWORK network;
NTSTATUS status;
BOOLEAN waitEvent = FALSE;
if (CnpNetworkShutdownEvent != NULL) {
IF_CNDBG(CN_DEBUG_INIT) {
CNPRINT(("[CNP] Cleaning up networks...\n"));
}
CnAcquireLock(&CnpNetworkListLock, &listIrql);
while (!IsListEmpty(&CnpNetworkList)) {
entry = CnpNetworkList.Flink;
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
network->Irql = DISPATCH_LEVEL;
CnpDeleteNetwork(network, listIrql);
//
// Both locks were released.
//
CnAcquireLock(&CnpNetworkListLock, &listIrql);
}
if (!IsListEmpty(&CnpDeletingNetworkList)) {
CnpIsNetworkShutdownPending = TRUE;
waitEvent = TRUE;
KeResetEvent(CnpNetworkShutdownEvent);
}
CnReleaseLock(&CnpNetworkListLock, listIrql);
if (waitEvent) {
IF_CNDBG(CN_DEBUG_INIT) {
CNPRINT(("[CNP] Network deletes are pending...\n"));
}
status = KeWaitForSingleObject(
CnpNetworkShutdownEvent,
Executive,
KernelMode,
FALSE, // not alertable
NULL // no timeout
);
CnAssert(status == STATUS_SUCCESS);
}
CnAssert(IsListEmpty(&CnpNetworkList));
CnAssert(IsListEmpty(&CnpDeletingNetworkList));
CnFreePool(CnpNetworkShutdownEvent); CnpNetworkShutdownEvent = NULL;
IF_CNDBG(CN_DEBUG_INIT) {
CNPRINT(("[CNP] Networks cleaned up.\n"));
}
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return;
}
NTSTATUS
CxRegisterNetwork(
CL_NETWORK_ID NetworkId,
ULONG Priority,
BOOLEAN Restricted
)
{
NTSTATUS status = STATUS_SUCCESS;
PLIST_ENTRY entry;
CN_IRQL listIrql;
PCNP_NETWORK network = NULL;
if (!CnpIsValidNetworkId(NetworkId)) {
return(STATUS_CLUSTER_INVALID_NETWORK);
}
//
// Allocate and initialize a network object.
//
network = CnAllocatePool(sizeof(CNP_NETWORK));
if (network == NULL) {
return(STATUS_INSUFFICIENT_RESOURCES);
}
RtlZeroMemory(network, sizeof(CNP_NETWORK));
CN_INIT_SIGNATURE(network, CNP_NETWORK_SIG);
network->RefCount = 1;
network->Id = NetworkId;
network->State = ClusnetNetworkStateOffline;
network->Priority = Priority;
if (Restricted) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Registering network %u as restricted\n", NetworkId));
}
network->Flags |= CNP_NET_FLAG_RESTRICTED;
}
else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Registering network %u as unrestricted\n", NetworkId));
}
}
CnpNetworkResetMcastReachableNodes(network);
CnInitializeLock(&(network->Lock), CNP_NETWORK_OBJECT_LOCK);
CnAcquireLock(&CnpNetworkListLock, &listIrql);
//
// Check if the specified network already exists.
//
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList;
entry = entry->Flink
)
{
PCNP_NETWORK oldNetwork = CONTAINING_RECORD(
entry,
CNP_NETWORK,
Linkage
);
CnAcquireLock(&(oldNetwork->Lock), &(oldNetwork->Irql));
if (NetworkId == oldNetwork->Id) {
CnReleaseLock(&(oldNetwork->Lock), oldNetwork->Irql);
CnReleaseLock(&CnpNetworkListLock, listIrql);
status = STATUS_CLUSTER_NETWORK_EXISTS;
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Network %u already exists\n", NetworkId));
}
goto error_exit;
}
CnReleaseLock(&(oldNetwork->Lock), oldNetwork->Irql);
}
InsertTailList(&CnpNetworkList, &(network->Linkage));
network->Flags |= CNP_NET_FLAG_MCASTSORTED;
CnReleaseLock(&CnpNetworkListLock, listIrql);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Registered network %u\n", NetworkId));
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(STATUS_SUCCESS);
error_exit:
if (network != NULL) {
CnFreePool(network);
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxRegisterNetwork
VOID
CxCancelDeregisterNetwork(
PDEVICE_OBJECT DeviceObject,
PIRP Irp
)
/*++
Routine Description:
Cancellation handler for DeregisterNetwork requests.
Return Value:
None.
Notes:
Called with cancel spinlock held.
Returns with cancel spinlock released.
--*/
{
PFILE_OBJECT fileObject;
CN_IRQL cancelIrql = Irp->CancelIrql;
PLIST_ENTRY entry;
PCNP_NETWORK network;
CnMarkIoCancelLockAcquired();
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Attempting to cancel DeregisterNetwork irp %p\n",
Irp
));
}
CnAssert(DeviceObject == CnDeviceObject);
fileObject = CnBeginCancelRoutine(Irp);
CnAcquireLockAtDpc(&CnpNetworkListLock);
CnReleaseCancelSpinLock(DISPATCH_LEVEL);
//
// We can only complete the irp if we can find it stashed in a
// deleting network object. The deleting network object could have
// been destroyed and the IRP completed before we acquired the
// CnpNetworkListLock.
//
for (entry = CnpDeletingNetworkList.Flink;
entry != &CnpDeletingNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
if (network->PendingDeleteIrp == Irp) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Found dereg irp on network %u\n",
network->Id
));
}
//
// Found the Irp. Now take it away and complete it.
//
network->PendingDeleteIrp = NULL;
CnReleaseLock(&CnpNetworkListLock, cancelIrql);
CnAcquireCancelSpinLock(&(Irp->CancelIrql));
CnEndCancelRoutine(fileObject);
CnCompletePendingRequest(Irp, STATUS_CANCELLED, 0);
//
// IoCancelSpinLock was released by CnCompletePendingRequest().
//
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return;
}
}
CnReleaseLock(&CnpNetworkListLock, cancelIrql);
CnAcquireCancelSpinLock(&cancelIrql);
CnEndCancelRoutine(fileObject);
CnReleaseCancelSpinLock(cancelIrql);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return;
} // CnpCancelApiDeregisterNetwork
NTSTATUS
CxDeregisterNetwork(
IN CL_NETWORK_ID NetworkId,
IN PIRP Irp,
IN PIO_STACK_LOCATION IrpSp
)
{
NTSTATUS status;
PLIST_ENTRY entry;
CN_IRQL irql;
PCNP_NETWORK network = NULL;
CnAcquireCancelSpinLock(&irql);
CnAcquireLockAtDpc(&CnpNetworkListLock);
status = CnMarkRequestPending(Irp, IrpSp, CxCancelDeregisterNetwork);
CnReleaseCancelSpinLock(DISPATCH_LEVEL);
if (status != STATUS_CANCELLED) {
CnAssert(status == STATUS_SUCCESS);
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
if (NetworkId == network->Id) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Deregistering network %u.\n",
NetworkId
));
}
//
// Save a pointer to pending irp. Note this is protected
// by the list lock, not the object lock.
//
network->PendingDeleteIrp = Irp;
CnpDeleteNetwork(network, irql);
//
// Both locks were released.
// Irp will be completed when the network is destroyed
// or the irp is cancelled.
//
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(STATUS_PENDING);
}
CnReleaseLockFromDpc(&(network->Lock));
}
CnReleaseLock(&CnpNetworkListLock, irql);
CnAcquireCancelSpinLock(&(Irp->CancelIrql));
CnCompletePendingRequest(Irp, STATUS_CLUSTER_NETWORK_NOT_FOUND, 0);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(STATUS_PENDING);
}
CnAssert(status == STATUS_CANCELLED);
CnReleaseLock(&CnpNetworkListLock, irql);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxDeregisterNetwork
NTSTATUS
CxOnlineNetwork(
IN CL_NETWORK_ID NetworkId,
IN PWCHAR TdiProviderName,
IN ULONG TdiProviderNameLength,
IN PTRANSPORT_ADDRESS TdiBindAddress,
IN ULONG TdiBindAddressLength,
IN PWCHAR AdapterName,
IN ULONG AdapterNameLength,
OUT PTDI_ADDRESS_INFO TdiBindAddressInfo,
IN ULONG TdiBindAddressInfoLength,
IN PIRP Irp OPTIONAL
)
/*++
Notes:
Each associated interface will be brought online when a heartbeat
is established for the target of the interface.
--*/
{
NTSTATUS status;
PCNP_NETWORK network;
OBJECT_ATTRIBUTES objectAttributes;
IO_STATUS_BLOCK iosb;
PFILE_FULL_EA_INFORMATION ea = NULL;
ULONG eaBufferLength;
HANDLE addressHandle = NULL;
PFILE_OBJECT addressFileObject = NULL;
PDEVICE_OBJECT addressDeviceObject = NULL;
BOOLEAN attached = FALSE;
UNICODE_STRING unicodeString;
TDI_REQUEST_KERNEL_QUERY_INFORMATION queryInfo;
PTDI_ADDRESS_INFO addressInfo;
//
// Allocate memory to hold the EA buffer we'll use to specify the
// transport address to NtCreateFile.
//
eaBufferLength = FIELD_OFFSET(FILE_FULL_EA_INFORMATION, EaName[0]) +
TDI_TRANSPORT_ADDRESS_LENGTH + 1 +
TdiBindAddressLength;
ea = CnAllocatePool(eaBufferLength);
if (ea == NULL) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] memory allocation of %u bytes failed.\n",
eaBufferLength
));
}
return(STATUS_INSUFFICIENT_RESOURCES);
}
//
// Initialize the EA using the network's transport information.
//
ea->NextEntryOffset = 0;
ea->Flags = 0;
ea->EaNameLength = TDI_TRANSPORT_ADDRESS_LENGTH;
ea->EaValueLength = (USHORT) TdiBindAddressLength;
RtlMoveMemory(
ea->EaName,
TdiTransportAddress,
ea->EaNameLength + 1
);
RtlMoveMemory(
&(ea->EaName[ea->EaNameLength + 1]),
TdiBindAddress,
TdiBindAddressLength
);
RtlInitUnicodeString(&unicodeString, TdiProviderName);
network = CnpFindNetwork(NetworkId);
if (network == NULL) {
CnFreePool(ea);
return(STATUS_CLUSTER_NETWORK_NOT_FOUND);
}
if (network->State != ClusnetNetworkStateOffline) {
CnReleaseLock(&(network->Lock), network->Irql);
CnFreePool(ea);
return(STATUS_CLUSTER_NETWORK_ALREADY_ONLINE);
}
CnAssert(network->DatagramHandle == NULL);
CnAssert(network->DatagramFileObject == NULL);
CnAssert(network->DatagramDeviceObject == NULL);
CnAssert(network->ActiveRefCount == 0);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Bringing network %u online...\n", NetworkId));
}
//
// Set the initial active refcount to 2. One reference will be removed
// when the network is successfully brought online. The other will be
// removed when the network is to be taken offline. Also increment the
// base refcount to account for the active refcount. Change to
// the online pending state.
//
network->ActiveRefCount = 2;
CnpReferenceNetwork(network);
network->State = ClusnetNetworkStateOnlinePending;
CnReleaseLock(&(network->Lock), network->Irql);
//
// Prepare for opening the address object.
//
InitializeObjectAttributes(
&objectAttributes,
&unicodeString,
OBJ_CASE_INSENSITIVE, // attributes
NULL,
NULL
);
//
// Attach to the system process so the handle we open will remain valid
// after the calling process goes away.
//
KeAttachProcess(CnSystemProcess);
attached = TRUE;
//
// Perform the actual open of the address object.
//
status = ZwCreateFile(
&addressHandle,
GENERIC_READ | GENERIC_WRITE | SYNCHRONIZE,
&objectAttributes,
&iosb, // returned status information.
0, // block size (unused).
0, // file attributes.
0, // not shareable
FILE_CREATE, // create disposition.
0, // create options.
ea,
eaBufferLength
);
CnFreePool(ea); ea = NULL;
if (status != STATUS_SUCCESS) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Failed to open address for network %u, status %lx.\n",
NetworkId,
status
));
}
goto error_exit;
}
//
// Get a pointer to the file object of the address.
//
status = ObReferenceObjectByHandle(
addressHandle,
0L, // DesiredAccess
NULL,
KernelMode,
&addressFileObject,
NULL
);
if (status != STATUS_SUCCESS) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Failed to reference address handle, status %lx.\n",
status
));
}
goto error_exit;
}
//
// Remember the device object to which we need to give requests for
// this address object. We can't just use the fileObject->DeviceObject
// pointer because there may be a device attached to the transport
// protocol.
//
addressDeviceObject = IoGetRelatedDeviceObject(
addressFileObject
);
//
// Adjust the StackSize of CdpDeviceObject so that we can pass CDP
// IRPs through for this network.
//
CnAdjustDeviceObjectStackSize(CdpDeviceObject, addressDeviceObject);
//
// Get the transport provider info
//
queryInfo.QueryType = TDI_QUERY_PROVIDER_INFO;
queryInfo.RequestConnectionInformation = NULL;
status = CnpIssueDeviceControl(
addressFileObject,
addressDeviceObject,
&queryInfo,
sizeof(queryInfo),
&(network->ProviderInfo),
sizeof(network->ProviderInfo),
TDI_QUERY_INFORMATION,
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Failed to get provider info, status %lx\n",
status
));
}
goto error_exit;
}
if (! ( network->ProviderInfo.ServiceFlags &
TDI_SERVICE_CONNECTIONLESS_MODE)
)
{
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Provider doesn't support datagrams!\n"
));
}
status = STATUS_CLUSTER_INVALID_NETWORK_PROVIDER;
goto error_exit;
}
//
// Get the address to which we were bound.
//
queryInfo.QueryType = TDI_QUERY_ADDRESS_INFO;
queryInfo.RequestConnectionInformation = NULL;
status = CnpIssueDeviceControl(
addressFileObject,
addressDeviceObject,
&queryInfo,
sizeof(queryInfo),
TdiBindAddressInfo,
TdiBindAddressInfoLength,
TDI_QUERY_INFORMATION,
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Failed to get address info, status %lx\n",
status
));
}
goto error_exit;
}
//
// Set up indication handlers on the address object. We are eligible
// to receive indications as soon as we do this.
//
status = CnpTdiSetEventHandler(
addressFileObject,
addressDeviceObject,
TDI_EVENT_ERROR,
CnpTdiErrorHandler,
network,
Irp
);
if ( !NT_SUCCESS(status) ) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Setting TDI_EVENT_ERROR failed: %lx\n",
status
));
}
goto error_exit;
}
status = CnpTdiSetEventHandler(
addressFileObject,
addressDeviceObject,
TDI_EVENT_RECEIVE_DATAGRAM,
CnpTdiReceiveDatagramHandler,
network,
Irp
);
if ( !NT_SUCCESS(status) ) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Setting TDI_EVENT_RECEIVE_DATAGRAM failed: %lx\n",
status
));
}
goto error_exit;
}
//
// We're done working with handles, so detach from the system process.
//
KeDetachProcess();
//
// Finish transition to online state. Note that an offline request
// could have been issued in the meantime.
//
CnAcquireLock(&(network->Lock), &(network->Irql));
network->DatagramHandle = addressHandle;
addressHandle = NULL;
network->DatagramFileObject = addressFileObject;
addressFileObject = NULL;
network->DatagramDeviceObject = addressDeviceObject;
addressDeviceObject = NULL;
//
// If an offline wasn't issued, change to the online state.
//
if (network->State == ClusnetNetworkStateOnlinePending) {
CnAssert(network->ActiveRefCount == 2);
network->State = ClusnetNetworkStateOnline;
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[CNP] Network %u is now online.\n", NetworkId));
}
CnReleaseLock(&(network->Lock), network->Irql);
//
// Bring all of the interfaces on this network online.
//
// The network can't be taken offline because we still hold
// the 2nd active reference.
//
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Bringing all interfaces on network %u online...\n",
network->Id
));
}
CnpWalkInterfacesOnNetwork(network, CnpOnlinePendingInterfaceWrapper);
//
// Position the network in the network list according to
// multicast reachability.
//
CnpSortMulticastNetwork(network, TRUE, NULL);
//
// Reacquire the lock to drop the active reference.
//
CnAcquireLock(&(network->Lock), &(network->Irql));
}
else {
//
// An offline was issued. It will take effect when we
// remove our 2nd active reference. The offline operation removed
// the first one. No send threads could have accessed this network
// yet because we never brought the associated interfaces online.
//
CnAssert(network->State == ClusnetNetworkStateOfflinePending);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] An offline request was issued on network %u during online pending.\n",
NetworkId
));
}
}
CnpActiveDereferenceNetwork(network);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(STATUS_SUCCESS);
error_exit:
if (addressFileObject != NULL) {
ObDereferenceObject(addressFileObject);
}
if (addressHandle != NULL) {
ZwClose(addressHandle);
}
KeDetachProcess();
CnAcquireLock(&(network->Lock), &(network->Irql));
if (network->State == ClusnetNetworkStateOnlinePending) {
//
// Remove our 2nd active reference and call the offline code.
// The offline function will release the network object lock.
//
CnAssert(network->ActiveRefCount == 2);
--(network->ActiveRefCount);
CnpOfflineNetwork(network);
}
else {
CnAssert(network->State == ClusnetNetworkStateOfflinePending);
//
// An offline was issued. It will take effect when we
// remove our 2nd active reference. The offline operation removed
// the first one. The dereference will release the network object
// lock.
//
CnAssert(network->State == ClusnetNetworkStateOfflinePending);
CnAssert(network->ActiveRefCount == 1);
CnpActiveDereferenceNetwork(network);
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxOnlineNetwork
VOID
CxCancelOfflineNetwork(
PDEVICE_OBJECT DeviceObject,
PIRP Irp
)
/*++
Routine Description:
Cancellation handler for OfflineNetwork requests.
Return Value:
None.
Notes:
Called with cancel spinlock held.
Returns with cancel spinlock released.
--*/
{
PFILE_OBJECT fileObject;
CN_IRQL cancelIrql = Irp->CancelIrql;
PLIST_ENTRY entry;
PCNP_NETWORK network;
PCNP_NETWORK offlineNetwork = NULL;
CnMarkIoCancelLockAcquired();
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Attempting to cancel OfflineNetwork irp %p\n",
Irp
));
}
CnAssert(DeviceObject == CnDeviceObject);
fileObject = CnBeginCancelRoutine(Irp);
CnAcquireLockAtDpc(&CnpNetworkListLock);
CnReleaseCancelSpinLock(DISPATCH_LEVEL);
//
// We can only complete the irp if we can find it stashed in a
// network object. The network object could have been destroyed
// and the IRP completed before we acquired the CnpNetworkListLock.
//
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
if (network->PendingOfflineIrp == Irp) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Found offline irp on network %u\n",
network->Id
));
}
network->PendingOfflineIrp = NULL;
offlineNetwork = network;
CnReleaseLockFromDpc(&(network->Lock));
break;
}
CnReleaseLockFromDpc(&(network->Lock));
}
if (offlineNetwork == NULL) {
for (entry = CnpDeletingNetworkList.Flink;
entry != &CnpDeletingNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
if (network->PendingOfflineIrp == Irp) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Found offline irp on network %u\n",
network->Id
));
}
network->PendingOfflineIrp = NULL;
offlineNetwork = network;
CnReleaseLockFromDpc(&(network->Lock));
break;
}
CnReleaseLockFromDpc(&(network->Lock));
}
}
CnReleaseLock(&CnpNetworkListLock, cancelIrql);
CnAcquireCancelSpinLock(&cancelIrql);
CnEndCancelRoutine(fileObject);
if (offlineNetwork != NULL) {
//
// Found the Irp. Now take it away and complete it.
// This releases the cancel spinlock
//
Irp->CancelIrql = cancelIrql;
CnCompletePendingRequest(Irp, STATUS_CANCELLED, 0);
}
else {
CnReleaseCancelSpinLock(cancelIrql);
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return;
} // CnpCancelApiOfflineNetwork
NTSTATUS
CxOfflineNetwork(
IN CL_NETWORK_ID NetworkId,
IN PIRP Irp,
IN PIO_STACK_LOCATION IrpSp
)
/*++
Notes:
--*/
{
PCNP_NETWORK network;
CN_IRQL irql;
NTSTATUS status;
CnAcquireCancelSpinLock(&irql);
CnAcquireLockAtDpc(&CnpNetworkListLock);
status = CnMarkRequestPending(Irp, IrpSp, CxCancelOfflineNetwork);
CnReleaseCancelSpinLock(DISPATCH_LEVEL);
if (status != STATUS_CANCELLED) {
CnAssert(status == STATUS_SUCCESS);
network = CnpLockedFindNetwork(NetworkId, irql);
//
// CnpNetworkListLock was released.
//
if (network != NULL) {
if (network->State >= ClusnetNetworkStateOnlinePending) {
network->PendingOfflineIrp = Irp;
CnpOfflineNetwork(network);
return(STATUS_PENDING);
}
CnReleaseLock(&(network->Lock), network->Irql);
status = STATUS_CLUSTER_NETWORK_ALREADY_OFFLINE;
}
else {
status = STATUS_CLUSTER_NETWORK_NOT_FOUND;
}
CnAcquireCancelSpinLock(&irql);
Irp->CancelIrql = irql;
CnCompletePendingRequest(Irp, status, 0);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(STATUS_PENDING);
}
CnAssert(status == STATUS_CANCELLED);
CnReleaseLock(&CnpNetworkListLock, irql);
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxOfflineNetwork
NTSTATUS
CxSetNetworkRestriction(
IN CL_NETWORK_ID NetworkId,
IN BOOLEAN Restricted,
IN ULONG NewPriority
)
{
NTSTATUS status;
PCNP_NETWORK network;
network = CnpFindNetwork(NetworkId);
if (network != NULL) {
if (Restricted) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Restricting network %u.\n",
network->Id
));
}
network->Flags |= CNP_NET_FLAG_RESTRICTED;
}
else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Unrestricting network %u.\n",
network->Id
));
}
network->Flags &= ~CNP_NET_FLAG_RESTRICTED;
if (NewPriority != 0) {
network->Priority = NewPriority;
}
}
//
// Reference the network so it can't go away while we
// reprioritize the associated interfaces.
//
CnpReferenceNetwork(network);
CnReleaseLock(&(network->Lock), network->Irql);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Recalculating priority for all interfaces on network %u ...\n",
network->Id
));
}
if (!Restricted) {
CnpWalkInterfacesOnNetwork(
network,
CnpRecalculateInterfacePriority
);
}
CnpWalkInterfacesOnNetwork(network, CnpReevaluateInterfaceRole);
//
// Reposition the network according to multicast reachability.
//
CnpSortMulticastNetwork(network, TRUE, NULL);
CnAcquireLock(&(network->Lock), &(network->Irql));
CnpDereferenceNetwork(network);
status = STATUS_SUCCESS;
}
else {
status = STATUS_CLUSTER_NETWORK_NOT_FOUND;
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxSetNetworkRestriction
NTSTATUS
CxSetNetworkPriority(
IN CL_NETWORK_ID NetworkId,
IN ULONG Priority
)
{
NTSTATUS status;
PCNP_NETWORK network;
if (Priority == 0) {
return(STATUS_INVALID_PARAMETER);
}
network = CnpFindNetwork(NetworkId);
if (network != NULL) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Network %u old priority %u, new priority %u.\n",
network->Id,
network->Priority,
Priority
));
}
network->Priority = Priority;
//
// Reference the network so it can't go away while we
// reprioritize the associated interfaces.
//
CnpReferenceNetwork(network);
CnReleaseLock(&(network->Lock), network->Irql);
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT((
"[CNP] Recalculating priority for all interfaces on network %u ...\n",
network->Id
));
}
CnpWalkInterfacesOnNetwork(network, CnpRecalculateInterfacePriority);
//
// Reposition the network according to multicast reachability.
//
CnpSortMulticastNetwork(network, TRUE, NULL);
CnAcquireLock(&(network->Lock), &(network->Irql));
CnpDereferenceNetwork(network);
status = STATUS_SUCCESS;
}
else {
status = STATUS_CLUSTER_NETWORK_NOT_FOUND;
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxSetNetworkPriority
NTSTATUS
CxGetNetworkPriority(
IN CL_NETWORK_ID NetworkId,
OUT PULONG Priority
)
{
NTSTATUS status;
PCNP_NETWORK network;
network = CnpFindNetwork(NetworkId);
if (network != NULL) {
*Priority = network->Priority;
CnReleaseLock(&(network->Lock), network->Irql);
status = STATUS_SUCCESS;
}
else {
status = STATUS_CLUSTER_NETWORK_NOT_FOUND;
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxGetNetworkPriority
NTSTATUS
CxGetNetworkState(
IN CL_NETWORK_ID NetworkId,
OUT PCLUSNET_NETWORK_STATE State
)
{
NTSTATUS status;
PCNP_NETWORK network;
network = CnpFindNetwork(NetworkId);
if (network != NULL) {
*State = network->State;
CnReleaseLock(&(network->Lock), network->Irql);
status = STATUS_SUCCESS;
}
else {
status = STATUS_CLUSTER_NETWORK_NOT_FOUND;
}
CnVerifyCpuLockMask(
0, // Required
0xFFFFFFFF, // Forbidden
0 // Maximum
);
return(status);
} // CxGetNetworkState
NTSTATUS
CxUnreserveClusnetEndpoint(
VOID
)
/*++
Routine Description:
Unreserves port number previously reserved with
CxUnreserveClusnetEndpoint.
CnResource should already be held when this routine
is called. Alternately, this routine is called without
CnResource held during unload of the clusnet driver.
Arguments:
None.
Return Value:
Status of TCP/IP ioctl.
--*/
{
HANDLE tcpHandle = (HANDLE) NULL;
TCP_RESERVE_PORT_RANGE portRange;
NTSTATUS status = STATUS_SUCCESS;
// Check if we have a port reserved
if (CnpReservedClusnetPort != 0) {
status = CnpOpenDevice(
DD_TCP_DEVICE_NAME,
&tcpHandle
);
if (NT_SUCCESS(status)) {
// TCP/IP interprets port ranges in host order
portRange.LowerRange = CnpReservedClusnetPort;
portRange.UpperRange = CnpReservedClusnetPort;
status = CnpZwDeviceControl(
tcpHandle,
IOCTL_TCP_UNRESERVE_PORT_RANGE,
&portRange,
sizeof(portRange),
NULL,
0
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Failed to unreserve "
"port %d: %lx\n",
CnpReservedClusnetPort, status));
}
} else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Unreserved "
"port %d.\n",
CnpReservedClusnetPort));
}
}
ZwClose(tcpHandle);
} else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Failed to open device %S, "
"status %lx\n",
DD_TCP_DEVICE_NAME, status));
}
}
CnpReservedClusnetPort = 0;
}
return status;
}
NTSTATUS
CxReserveClusnetEndpoint(
IN USHORT Port
)
/*++
Routine Description:
Reserves assigned clusnet endpoint port number so that
the TCP/IP driver will not hand it out to applications
requesting a wildcard port.
Arguments:
Port - port number to reserve, in host byte-order format
Return Value:
Status of TCP/IP ioctl.
--*/
{
NTSTATUS status = STATUS_SUCCESS;
HANDLE tcpHandle = (HANDLE) NULL;
TCP_RESERVE_PORT_RANGE portRange;
// Check for invalid port number 0
if (Port == 0) {
return STATUS_INVALID_PARAMETER;
}
// Check if we already have a port reserved.
if (CnpReservedClusnetPort != 0
&& CnpReservedClusnetPort != Port) {
status = CxUnreserveClusnetEndpoint();
}
if (CnpReservedClusnetPort == 0) {
// Reserve Port with the TCP/IP driver.
status = CnpOpenDevice(
DD_TCP_DEVICE_NAME,
&tcpHandle
);
if (NT_SUCCESS(status)) {
// TCP/IP interprets port ranges in host order
portRange.LowerRange = Port;
portRange.UpperRange = Port;
status = CnpZwDeviceControl(
tcpHandle,
IOCTL_TCP_RESERVE_PORT_RANGE,
&portRange,
sizeof(portRange),
NULL,
0
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Failed to reserve "
"port %d: %lx\n",
Port, status));
}
} else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Reserved "
"port %d.\n",
Port));
}
CnpReservedClusnetPort = Port;
}
ZwClose(tcpHandle);
} else {
IF_CNDBG(CN_DEBUG_CONFIG) {
CNPRINT(("[Clusnet] Failed to open device %S, "
"status %lx\n",
DD_TCP_DEVICE_NAME, status));
}
}
}
return status;
} // CxReserveClusnetEndpoint
NTSTATUS
CxConfigureMulticast(
IN CL_NETWORK_ID NetworkId,
IN ULONG MulticastNetworkBrand,
IN PTRANSPORT_ADDRESS TdiMcastBindAddress,
IN ULONG TdiMcastBindAddressLength,
IN PVOID Key,
IN ULONG KeyLength,
IN PIRP Irp
)
/*++
Routine Description:
Configures a network for multicast.
Notes:
The network multicast flag is turned off at the
beginning of this routine to prevent multicasts
during the transition. If the routine does not
complete successfully, the multicast flag is
purposely left cleared.
--*/
{
NTSTATUS status;
KIRQL irql;
PLIST_ENTRY entry;
PCNP_NETWORK network = NULL;
BOOLEAN networkLocked = FALSE;
BOOLEAN mcastEnabled = FALSE;
TDI_REQUEST_KERNEL_QUERY_INFORMATION queryInfo;
PTDI_ADDRESS_INFO addressInfo;
HANDLE networkHandle;
PFILE_OBJECT networkFileObject;
PDEVICE_OBJECT networkDeviceObject;
PCNP_MULTICAST_GROUP group = NULL;
PCNP_MULTICAST_GROUP delGroup = NULL;
PCNP_MULTICAST_GROUP currGroup = NULL;
PCNP_MULTICAST_GROUP prevGroup = NULL;
BOOLEAN prevGroupMatch = FALSE;
UCHAR addressInfoBuffer[FIELD_OFFSET(TDI_ADDRESS_INFO, Address) +
sizeof(TA_IP_ADDRESS)] = {0};
//
// Validate multicast bind address parameter. Even if this
// request only leaves a group, the multicast address data
// structure must be provided.
//
if (TdiMcastBindAddressLength != sizeof(TA_IP_ADDRESS)) {
return (STATUS_INVALID_PARAMETER);
}
//
// Acquire the lock on the local node, but go through the
// node table to be extra paranoid.
//
CnAcquireLock(&CnpNodeTableLock, &irql);
if (CnpLocalNode != NULL) {
CnAcquireLockAtDpc(&(CnpLocalNode->Lock));
CnReleaseLockFromDpc(&CnpNodeTableLock);
CnpLocalNode->Irql = irql;
//
// Find the network object in the network object table.
//
CnAcquireLockAtDpc(&CnpNetworkListLock);
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
if (NetworkId == network->Id) {
//
// We now hold locks on the node, the network list,
// and the network.
//
// Verify the network state. Then take an active
// reference so that it doesn't disappear while
// we're working with it.
//
if ((network->State < ClusnetNetworkStateOnline) ||
(!CnpActiveReferenceNetwork(network))
) {
CnReleaseLockFromDpc(&(network->Lock));
CnReleaseLockFromDpc(&CnpNetworkListLock);
CnReleaseLock(&(CnpLocalNode->Lock), CnpLocalNode->Irql);
return(STATUS_CLUSTER_INVALID_NETWORK);
}
//
// Clear the reachable set for this network.
//
CnpMulticastChangeNodeReachabilityLocked(
network,
CnpLocalNode,
FALSE,
TRUE,
NULL
);
//
// Remember whether the network was multicast-capable. Then
// clear the multicast capable flag so that we don't try to
// send multicasts during the transition period.
//
mcastEnabled = (BOOLEAN) CnpIsNetworkMulticastCapable(network);
network->Flags &= ~CNP_NET_FLAG_MULTICAST;
networkHandle = network->DatagramHandle;
networkFileObject = network->DatagramFileObject;
networkDeviceObject = network->DatagramDeviceObject;
currGroup = network->CurrentMcastGroup;
if (currGroup != NULL) {
CnpReferenceMulticastGroup(currGroup);
}
prevGroup = network->PreviousMcastGroup;
if (prevGroup != NULL) {
CnpReferenceMulticastGroup(prevGroup);
}
//
// Release the network lock.
//
CnReleaseLockFromDpc(&(network->Lock));
networkLocked = FALSE;
//
// Break out of the network search.
//
break;
} else {
CnReleaseLockFromDpc(&(network->Lock));
network = NULL;
}
}
//
// Release the network list lock.
//
CnReleaseLockFromDpc(&CnpNetworkListLock);
//
// Release the local node lock.
//
CnReleaseLock(&(CnpLocalNode->Lock), CnpLocalNode->Irql);
} else {
CnReleaseLock(&CnpNodeTableLock, irql);
CnTrace(CNP_NET_DETAIL, CnpTraceMcastPreConfigNoHost,
"[CNP] Cannot configure multicast for network %u "
"because local host not found.",
NetworkId
);
return(STATUS_HOST_UNREACHABLE);
}
//
// Verify that we found the network.
//
if (network == NULL) {
return (STATUS_CLUSTER_NETWORK_NOT_FOUND);
}
//
// Allocate a multicast group data structure with the
// new configuration parameters.
//
status = CnpAllocateMulticastGroup(
MulticastNetworkBrand,
TdiMcastBindAddress,
TdiMcastBindAddressLength,
Key,
KeyLength,
&group
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Failed to allocate mcast group, "
"status %lx\n",
status
));
}
goto error_exit;
}
//
// Get the local interface address.
//
addressInfo = (PTDI_ADDRESS_INFO) &addressInfoBuffer[0];
queryInfo.QueryType = TDI_QUERY_ADDRESS_INFO;
queryInfo.RequestConnectionInformation = NULL;
status = CnpIssueDeviceControl(
networkFileObject,
networkDeviceObject,
&queryInfo,
sizeof(queryInfo),
addressInfo,
sizeof(addressInfoBuffer),
TDI_QUERY_INFORMATION,
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Failed to get address info, status %lx\n",
status
));
}
goto error_exit;
}
//
// Determine if the multicast bind address is valid. If not,
// we are disabling.
//
if (CnpIsIPv4McastTransportAddress(TdiMcastBindAddress)) {
// temp vars for logging
ULONG mcastBindIp;
ULONG ifBindIp;
mcastBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)TdiMcastBindAddress)
->Address[0].Address[0].in_addr)
)
);
ifBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)&(addressInfo->Address))
->Address[0].Address[0].in_addr)
)
);
//
// We are trying to join a new multicast group. Fail
// immediately if there is no key.
//
if (KeyLength == 0) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Cannot configure new multicast group "
"without key.\n"
));
}
status = STATUS_INVALID_PARAMETER;
goto error_exit;
}
//
// Configure basic multicast settings if not done
// previously (though this call is idempotent).
//
if (!mcastEnabled) {
status = CnpConfigureBasicMulticastSettings(
networkHandle,
networkFileObject,
networkDeviceObject,
addressInfo,
1, // ttl
0, // disable loopback
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Failed to configure basic "
"multicast settings, status %lx\n",
status
));
}
goto error_exit;
}
}
//
// Add the group address if we are not already in
// this multicast group (e.g. the multicast bind
// address is the same as the current group or
// previous group).
//
if (prevGroup != NULL &&
CnpIsIPv4McastSameGroup(
prevGroup->McastTdiAddress,
TdiMcastBindAddress
)
) {
prevGroupMatch = TRUE;
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] New mcast address matches "
"previous mcast address.\n"));
}
CnTrace(CNP_NET_DETAIL, CnpTraceMcastNewPrevGroupMatch,
"[CNP] New multicast group %!ipaddr! "
"for interface %!ipaddr!, network id %d, "
"matches previous group.",
mcastBindIp,
ifBindIp,
NetworkId
);
}
else if (currGroup != NULL &&
CnpIsIPv4McastSameGroup(
currGroup->McastTdiAddress,
TdiMcastBindAddress
)
) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] New mcast address matches "
"current mcast address.\n"));
}
CnTrace(CNP_NET_DETAIL, CnpTraceMcastNewCurrGroupMatch,
"[CNP] New multicast group %!ipaddr! "
"for interface %!ipaddr!, network id %d, "
"matches current group.",
mcastBindIp,
ifBindIp,
NetworkId
);
} else {
status = CnpAddRemoveMulticastAddress(
networkHandle,
networkFileObject,
networkDeviceObject,
addressInfo,
TdiMcastBindAddress,
AO_OPTION_ADD_MCAST,
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Failed to add mcast address, "
"status %lx\n",
status
));
}
CnTrace(CNP_NET_DETAIL, CnpTraceMcastAddGroupFailed,
"[CNP] Failed to add multicast group %!ipaddr! to "
"interface %!ipaddr! network id %d, "
"status %!status!.",
mcastBindIp,
ifBindIp,
NetworkId,
status
);
goto error_exit;
} else {
CnTrace(CNP_NET_DETAIL, CnpTraceMcastAddGroup,
"[CNP] Added multicast group %!ipaddr! to "
"interface %!ipaddr!, network id %d.",
mcastBindIp,
ifBindIp,
NetworkId
);
}
}
}
//
// Leave membership for previous group if
// - the previous group does not match the new group, AND
// - the previous group does not match the current group
//
if (!prevGroupMatch &&
prevGroup != NULL &&
CnpIsIPv4McastTransportAddress(prevGroup->McastTdiAddress)) {
// temp vars for logging
ULONG mcastBindIp;
ULONG ifBindIp;
mcastBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)prevGroup->McastTdiAddress)
->Address[0].Address[0].in_addr)
)
);
ifBindIp = *((ULONG UNALIGNED *)
(&(((PTA_IP_ADDRESS)&(addressInfo->Address))
->Address[0].Address[0].in_addr)
)
);
if (!CnpIsIPv4McastSameGroup(
prevGroup->McastTdiAddress,
currGroup->McastTdiAddress
)) {
status = CnpAddRemoveMulticastAddress(
networkHandle,
networkFileObject,
networkDeviceObject,
addressInfo,
prevGroup->McastTdiAddress,
AO_OPTION_DEL_MCAST,
Irp
);
if (!NT_SUCCESS(status)) {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT((
"[CNP] Failed to leave mcast group, "
"IF %d.%d.%d.%d, mcast addr %d.%d.%d.%d, "
"status %lx.\n",
CnpIpAddrPrintArgs(ifBindIp),
CnpIpAddrPrintArgs(mcastBindIp),
status
));
}
CnTrace(CNP_NET_DETAIL, CnpTraceMcastLeaveGroupFailed,
"[CNP] Failed to leave multicast group %!ipaddr! on "
"interface %!ipaddr!, network id %d, "
"status %!status!. Continuing anyway.",
mcastBindIp,
ifBindIp,
NetworkId,
status
);
// not considered a fatal error
status = STATUS_SUCCESS;
} else {
CnTrace(CNP_NET_DETAIL, CnpTraceMcastLeaveGroup,
"[CNP] Left multicast group %!ipaddr! on "
"interface %!ipaddr!, network id %d.",
mcastBindIp,
ifBindIp,
NetworkId
);
}
} else {
IF_CNDBG(CN_DEBUG_NETOBJ) {
CNPRINT(("[CNP] Prev mcast address matches "
"current mcast address.\n"));
}
CnTrace(CNP_NET_DETAIL, CnpTraceMcastCurrPrevGroupMatch,
"[CNP] Previous multicast group %!ipaddr! "
"on interface %!ipaddr!, network id %d, "
"matches current group. Not leaving.",
mcastBindIp,
ifBindIp,
NetworkId
);
}
}
//
// Reacquire the network lock to make changes
// to the network object data structure, including
// shifting the multicast group data structures and
// turning on the multicast flag. The multicast flag
// was turned off earlier in this routine before
// starting the transition. It is only re-enabled if
// the new multicast bind address is a valid multicast
// address.
//
CnAcquireLock(&(network->Lock), &(network->Irql));
networkLocked = TRUE;
delGroup = network->PreviousMcastGroup;
network->PreviousMcastGroup = network->CurrentMcastGroup;
network->CurrentMcastGroup = group;
group = NULL;
if (CnpIsIPv4McastTransportAddress(TdiMcastBindAddress)) {
network->Flags |= CNP_NET_FLAG_MULTICAST;
}
CnReleaseLock(&(network->Lock), network->Irql);
networkLocked = FALSE;
//
// Transition into new multicast group, if appropriate.
//
if (CnpIsIPv4McastTransportAddress(TdiMcastBindAddress)) {
//
// Switch all outgoing heartbeats on this node to
// unicast with discovery.
//
CnpWalkInterfacesOnNetwork(
network,
CnpStartInterfaceMcastTransition
);
}
error_exit:
if (networkLocked) {
CnReleaseLock(&(network->Lock), network->Irql);
networkLocked = FALSE;
}
//
// Reposition the network according to multicast reachability.
//
CnpSortMulticastNetwork(network, TRUE, NULL);
CnAcquireLock(&(network->Lock), &(network->Irql));
CnpActiveDereferenceNetwork(network);
if (group != NULL) {
CnpDereferenceMulticastGroup(group);
}
if (currGroup != NULL) {
CnpDereferenceMulticastGroup(currGroup);
}
if (prevGroup != NULL) {
CnpDereferenceMulticastGroup(prevGroup);
}
if (delGroup != NULL) {
CnpDereferenceMulticastGroup(delGroup);
}
return(status);
} // CxConfigureMulticast
BOOLEAN
CnpSortMulticastNetwork(
IN PCNP_NETWORK Network,
IN BOOLEAN RaiseEvent,
OUT CX_CLUSTERSCREEN * McastReachableNodes OPTIONAL
)
/*++
Routine Description:
Wrapper for CnpSortMulticastNetworkLocked.
Return value:
TRUE if reachable node set changed
Notes:
Acquires and releases CnpNetworkListLock.
--*/
{
KIRQL irql;
BOOLEAN setChanged = FALSE;
CnVerifyCpuLockMask(
0, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
CnAcquireLock(&CnpNetworkListLock, &irql);
setChanged = CnpSortMulticastNetworkLocked(
Network,
RaiseEvent,
McastReachableNodes
);
CnReleaseLock(&CnpNetworkListLock, irql);
CnVerifyCpuLockMask(
0, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
return(setChanged);
} // CnpSortMulticastNetwork
BOOLEAN
CnpMulticastChangeNodeReachability(
IN PCNP_NETWORK Network,
IN PCNP_NODE Node,
IN BOOLEAN Reachable,
IN BOOLEAN RaiseEvent,
OUT CX_CLUSTERSCREEN * NewMcastReachableNodes
)
/*++
Routine Description:
Changes the multicast reachability state of Node
on Network.
If the set of reachable nodes changes, returns
the new screen through NewMcastReachableNodes.
Return value:
TRUE if set of reachable nodes changes.
Notes:
Called and returns with node lock held.
--*/
{
KIRQL irql;
BOOLEAN setChanged = FALSE;
CnVerifyCpuLockMask(
CNP_NODE_OBJECT_LOCK, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
CnAcquireLock(&CnpNetworkListLock, &irql);
setChanged = CnpMulticastChangeNodeReachabilityLocked(
Network,
Node,
Reachable,
RaiseEvent,
NewMcastReachableNodes
);
CnReleaseLock(&CnpNetworkListLock, irql);
CnVerifyCpuLockMask(
CNP_NODE_OBJECT_LOCK, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
return(setChanged);
} // CnpMulticastChangeNodeReachability
PCNP_NETWORK
CnpGetBestMulticastNetwork(
VOID
)
/*++
Routine Description:
Returns network object that currently has best
node reachability.
Return value:
Best network object, or NULL if there are no
internal multicast networks.
Notes:
Must not be called with network list lock held.
Returns with network locked (if found).
--*/
{
PCNP_NETWORK network = NULL;
KIRQL listIrql;
KIRQL networkIrql;
CnVerifyCpuLockMask(
0, // required
(CNP_NETWORK_LIST_LOCK | CNP_NETWORK_OBJECT_LOCK), // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
CnAcquireLock(&CnpNetworkListLock, &listIrql);
if (!IsListEmpty(&CnpNetworkList)) {
network = CONTAINING_RECORD(
CnpNetworkList.Flink,
CNP_NETWORK,
Linkage
);
CnAcquireLock(&(network->Lock), &networkIrql);
if (CnpIsInternalMulticastNetwork(network)) {
CnReleaseLock(&CnpNetworkListLock, networkIrql);
network->Irql = listIrql;
CnVerifyCpuLockMask(
CNP_NETWORK_OBJECT_LOCK, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NETWORK_OBJECT_LOCK_MAX // max
);
} else {
CnReleaseLock(&(network->Lock), networkIrql);
network = NULL;
}
}
if (network == NULL) {
CnReleaseLock(&CnpNetworkListLock, listIrql);
CnVerifyCpuLockMask(
0, // required
(CNP_NETWORK_LIST_LOCK | CNP_NETWORK_OBJECT_LOCK), // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
}
return(network);
} // CnpGetBestMulticastNetwork
NTSTATUS
CxGetMulticastReachableSet(
IN CL_NETWORK_ID NetworkId,
OUT ULONG * NodeScreen
)
/*++
Routine Description:
Queries multicast reachable set for specified network.
The multicast reachable set is protected by the network
list lock.
--*/
{
KIRQL irql;
PLIST_ENTRY entry;
PCNP_NETWORK network;
CX_CLUSTERSCREEN nodeScreen;
BOOLEAN found = FALSE;
CnVerifyCpuLockMask(
0, // required
CNP_NETWORK_LIST_LOCK, // forbidden
CNP_NODE_OBJECT_LOCK_MAX // max
);
CnAcquireLock(&CnpNetworkListLock, &irql);
for (entry = CnpNetworkList.Flink;
entry != &CnpNetworkList && !found;
entry = entry->Flink
)
{
network = CONTAINING_RECORD(entry, CNP_NETWORK, Linkage);
CnAcquireLockAtDpc(&(network->Lock));
if (NetworkId == network->Id) {
nodeScreen = network->McastReachableNodes;
found = TRUE;
}
CnReleaseLockFromDpc(&(network->Lock));
}
CnReleaseLock(&CnpNetworkListLock, irql);
if (!found) {
return(STATUS_CLUSTER_NETWORK_NOT_FOUND);
} else {
*NodeScreen = nodeScreen.UlongScreen;
return(STATUS_SUCCESS);
}
} // CxGetMulticastReachableSet