able/windows-nt
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity
windows-nt/Source/XPSP1/NT/base/cluster/service/fm/worker.c

758 lines
24 KiB
C
Raw Normal View History

Add source files
2020-09-26 03:20:57 -05:00
/*++
Copyright (c) 1996 Microsoft Corporation
Module Name:
worker.c
Abstract:
Failover Manager worker thread.
Author:
Mike Massa (mikemas) 12-Mar-1996
Revision History:
--*/
#define UNICODE 1
#include "fmp.h"
#define LOG_MODULE WORKER
CL_QUEUE FmpWorkQueue;
//
// Local Data
//
HANDLE FmpWorkerThreadHandle = NULL;
//
// Forward routines
//
BOOL
FmpAddNodeGroupCallback(
IN PVOID Context1,
IN PVOID Context2,
IN PVOID Object,
IN LPCWSTR Name
);
//
// Local routines
//
DWORD
FmpWorkerThread(
IN LPVOID Ignored
)
{
DWORD status;
PLIST_ENTRY entry;
PWORK_ITEM workItem;
DWORD running = TRUE;
PFM_RESOURCE resource;
PFMP_POSSIBLE_NODE possibleNode;
DWORD i;
ClRtlLogPrint(LOG_NOISE,"[FM] Worker thread running\n");
while ( running ) {
//
// Check the FM work queue for work items
//
entry = ClRtlRemoveHeadQueue(&FmpWorkQueue);
if ( entry == NULL ) {
return(ERROR_SUCCESS);
}
workItem = CONTAINING_RECORD(entry,
WORK_ITEM,
ListEntry);
//
// FM no longer cares about node up events, make sure
// we aren't getting any queued.
//
switch ( workItem->Event ) {
case FM_EVENT_SHUTDOWN:
ClRtlLogPrint(LOG_NOISE,"[FM] WorkerThread terminating...\n");
running = FALSE;
break;
case FM_EVENT_RESOURCE_ADDED:
resource = workItem->Context1;
if ( resource->Monitor == NULL ) {
FmpAcquireLocalResourceLock( resource );
//
// Chittur Subbaraman (chitturs) - 8/12/99
//
// Make sure the resource is not marked for delete.
//
if ( IS_VALID_FM_RESOURCE( resource ) )
{
FmpInitializeResource( resource, TRUE );
}
FmpReleaseLocalResourceLock( resource );
}
OmDereferenceObject( resource );
break;
case FM_EVENT_RESOURCE_DELETED:
//
// Tell the resource monitor to cleanup the resource.
//
resource = workItem->Context1;
FmpAcquireLocalResourceLock( resource );
// Now that no remaining resource depends on this resource and
// this resource does not depend on any other resources, we can
// terminate it in the resource monitor if the resource is not
// already offline or failed
//
if ( (resource->Group->OwnerNode == NmLocalNode) &&
((resource->State != ClusterResourceOffline) &&
(resource->State != ClusterResourceFailed))) {
FmpRmTerminateResource(resource);
}
status = FmpRmCloseResource(resource);
ClRtlLogPrint( LOG_NOISE,
"[FM] WorkItem, delete resource <%1!ws!> status %2!u!\n",
OmObjectName(resource),
status );
FmpReleaseLocalResourceLock( resource );
OmDereferenceObject(resource);
break;
case FM_EVENT_GROUP_FAILED:
FmpHandleGroupFailure( workItem->Context1 );
break;
case FM_EVENT_NODE_ADDED:
//
// We need to add this node to every resource's possible owners
// list and to each group's preferred owners list.
//
OmEnumObjects( ObjectTypeGroup,
FmpAddNodeGroupCallback,
workItem->Context1,
NULL );
break;
case FM_EVENT_NODE_EVICTED:
//
// Enumerate all the resources types to remove any PossibleNode references.
//
OmEnumObjects(ObjectTypeResType,
FmpEnumResTypeNodeEvict,
workItem->Context1,
NULL);
//
// Enumerate all the resources to remove any PossibleNode references.
//
OmEnumObjects(ObjectTypeResource,
FmpEnumResourceNodeEvict,
workItem->Context1,
NULL);
//
// Enumerate all the groups to remove any PreferredNode references
//
OmEnumObjects(ObjectTypeGroup,
FmpEnumGroupNodeEvict,
workItem->Context1,
NULL);
//Now dereference the object
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_CLUSTER_PROPERTY_CHANGE:
//this is for the cluster name change
FmpClusterEventPropHandler((PFM_RESOURCE)workItem->Context1);
//Now dereference the object
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_RESOURCE_CHANGE:
// this is for a Add/Remove possible node request
possibleNode = workItem->Context1;
if ( possibleNode == NULL ) {
break;
}
FmpRmResourceControl( possibleNode->Resource,
possibleNode->ControlCode,
(PUCHAR)OmObjectName(possibleNode->Node),
((lstrlenW(OmObjectName(possibleNode->Node)) + 1) * sizeof(WCHAR)),
NULL,
0,
NULL,
NULL );
// ignore status
OmDereferenceObject( possibleNode->Resource );
OmDereferenceObject( possibleNode->Node );
LocalFree( possibleNode );
break;
case FM_EVENT_RESOURCE_PROPERTY_CHANGE:
//
// Generate a cluster wide event notification for this event.
//
ClusterWideEvent(
CLUSTER_EVENT_RESOURCE_PROPERTY_CHANGE,
workItem->Context1 // Resource
);
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_RES_RESOURCE_TRANSITION:
FmpHandleResourceTransition(workItem->Context1, workItem->Context2);
OmDereferenceObject(workItem->Context1);
break;
case FM_EVENT_RES_RESOURCE_FAILED:
FmpProcessResourceEvents(workItem->Context1, ClusterResourceFailed,
workItem->Context2);
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_RES_RETRY_TIMER:
resource= (PFM_RESOURCE)workItem->Context1;
//Remove any pending watchdog timer
if (resource->hTimer)
{
RemoveTimerActivity(resource->hTimer);
resource->hTimer = NULL;
}
FmpAcquireLocalResourceLock(resource);
// Check if this resource was deleted in the meanwhile,
// or is not in failed state
if( ( IS_VALID_FM_RESOURCE( resource ) ) &&
( resource->State == ClusterResourceFailed ) &&
( resource->PersistentState == ClusterResourceOnline ) )
{
// Check if we are the owner, if not ignore it
if ( resource->Group->OwnerNode == NmLocalNode )
{
FmpProcessResourceEvents(resource,
ClusterResourceFailed,
ClusterResourceOnline);
}
}
FmpReleaseLocalResourceLock(resource);
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_INTERNAL_PROP_GROUP_STATE:
FmpPropagateGroupState(workItem->Context1);
OmDereferenceObject( workItem->Context1 );
break;
case FM_EVENT_INTERNAL_RETRY_ONLINE:
{
PFM_RESLIST_ONLINE_RETRY_INFO pFmOnlineRetryInfo;
RemoveTimerActivity((HANDLE)workItem->Context2);
pFmOnlineRetryInfo= workItem->Context1;
CL_ASSERT(pFmOnlineRetryInfo);
ClRtlLogPrint(LOG_NOISE,
"[FM] FmpWorkerThread, retrying to online resourcelist\r\n");
FmpOnlineResourceFromList(&(pFmOnlineRetryInfo->ResourceEnum));
//Free memory
for (i =0; i< pFmOnlineRetryInfo->ResourceEnum.EntryCount; i++)
LocalFree( pFmOnlineRetryInfo->ResourceEnum.Entry[i].Id );
LocalFree(pFmOnlineRetryInfo);
break;
}
case FM_EVENT_INTERNAL_RESOURCE_CHANGE_PARAMS:
{
BOOL bIsValidRes = TRUE;
//
// Now tell the resource monitor about the changes.
//
status = ERROR_SUCCESS;
resource = (PFM_RESOURCE)workItem->Context1;
FmpAcquireLocalResourceLock( resource );
//
// Chittur Subbaraman (chitturs) - 8/12/99
//
// Check whether the resource is marked for delete.
//
if ( !IS_VALID_FM_RESOURCE( resource ) )
{
bIsValidRes = FALSE;
}
FmpReleaseLocalResourceLock( resource );
if( bIsValidRes )
{
status = FmpRmChangeResourceParams( resource );
}
if ( status != ERROR_SUCCESS )
{
ClRtlLogPrint(LOG_UNUSUAL,
"[FM] FmpWorkerThread, failed to change resource "
"parameters for %1!ws!, error %2!u!.\n",
OmObjectId(resource),
status );
}
OmDereferenceObject(resource);
break;
}
case FM_EVENT_INTERNAL_ONLINE_GROUPLIST:
{
PGROUP_ENUM pGroupList = NULL;
ClRtlLogPrint(LOG_NOISE,
"[FM] FmpWorkerThread : Processing Node Down Group List\n");
pGroupList = workItem->Context1;
FmpOnlineGroupList(pGroupList, TRUE);
FmpDeleteEnum(pGroupList);
break;
}
case FM_EVENT_RESOURCE_NAME_CHANGE:
{
//
// Chittur Subbaraman (chitturs) - 6/29/99
//
// Added this new event to handle resource name change
// notifications to resource DLLs.
//
PFM_RES_CHANGE_NAME pResChangeName = NULL;
DWORD dwStatus = ERROR_SUCCESS;
pResChangeName = ( PFM_RES_CHANGE_NAME ) workItem->Context1;
dwStatus = FmpRmResourceControl( pResChangeName->pResource,
CLUSCTL_RESOURCE_SET_NAME,
(PUCHAR) pResChangeName->szNewResourceName,
((lstrlenW(pResChangeName->szNewResourceName) + 1) * sizeof(WCHAR)),
NULL,
0,
NULL,
NULL );
ClRtlLogPrint(LOG_NOISE,
"[FM] Worker thread handling FM_EVENT_RESOURCE_NAME_CHANGE event - FmpRmResourceControl returns %1!u! for resource %2!ws!\n",
dwStatus,
OmObjectId(pResChangeName->pResource));
OmDereferenceObject( pResChangeName->pResource );
LocalFree( pResChangeName );
break;
}
default:
ClRtlLogPrint(LOG_NOISE,
"[FM] WorkerThread, unrecognized event %1!u!\n",
workItem->Event);
}
//
// Free the work item.
//
LocalFree( workItem );
}
return(ERROR_SUCCESS);
} // FmpWorkerThread
VOID
FmpProcessResourceEvents(
IN PFM_RESOURCE pResource,
IN CLUSTER_RESOURCE_STATE NewState,
IN CLUSTER_RESOURCE_STATE OldState
)
/*++
Routine Description:
Arguments:
Return Value:
Comments: This should not call PropagateResourceState(). FmpProcessResourceEvents
acquires the group lock. The quorum resource state must be propagated without holding
the group lock. FmpPropagateResourceState() must be called by FmpHandleResourceTransition.
There is a slight window between when the event is received in FmpHandleResourceTransition()
and when the actions corresponding to those are carried out in FmpProcessResourceEvents().
In this window, another opposing action like offline/online might occur. But we dont
worry about it since if there are waiting resources on this resource, those actions
are not carried out.
--*/
{
DWORD Status;
BOOL bQuoChangeLockHeld = FALSE;
CL_ASSERT(pResource != NULL);
ChkFMState:
if (!FmpFMGroupsInited)
{
DWORD dwRetryCount = 50;
ACQUIRE_SHARED_LOCK(gQuoChangeLock);
//FmFormNewClusterPhaseProcessing is in progress
if (FmpFMFormPhaseProcessing)
{
ClRtlLogPrint(LOG_CRITICAL,
"[FM] FmpProcessResourceEvents, resource notification from quorum resource during phase processing,sleep and retry\n");
RELEASE_LOCK(gQuoChangeLock);
Sleep(500);
if (dwRetryCount--)
goto ChkFMState;
else
{
ClRtlLogPrint(LOG_CRITICAL,
"[FM] FmpProcessResourceEvents, waited for too long\n");
//terminate the process
CL_ASSERT(FALSE);
}
}
else
{
bQuoChangeLockHeld = TRUE;
}
//this can only come from the quorum resource
CL_ASSERT(pResource->QuorumResource);
}
FmpAcquireLocalResourceLock( pResource );
//
// Chittur Subbaraman (chitturs) - 8/12/99
//
// First check whether the resource has been marked for deletion. If
// so, don't do anything. Note that this function is called from
// the worker thread AFTER FmpHandleResourceTransition has propagated
// the failed state of the resource. Now, after the propagation has
// occurred, a client is free to delete the resource. So, when the
// worker thread makes this function call, we need to check whether
// the resource is deleted and reject the call. Note that this function
// holds the resource lock on the owner node of the resource and so
// is serialized with the FmDeleteResource call which is also executed
// on the owner node of the resource with the lock held. So, the
// following check will give us a consistent result.
//
if ( !IS_VALID_FM_RESOURCE( pResource ) )
{
ClRtlLogPrint(LOG_NOISE,
"[FM] FmpProcessResourceEvents: Resource %1!ws! has already been deleted, returning...\n",
OmObjectId(pResource));
goto FnExit;
}
switch(NewState){
case ClusterResourceFailed:
//check the old resource state
if (OldState == ClusterResourceOnline)
{
FmpHandleResourceFailure(pResource);
}
else if ((OldState == ClusterResourceOffline) ||
(OldState == ClusterResourceOfflinePending))
{
FmpTerminateResource(pResource);
if ( pResource->Group->OwnerNode == NmLocalNode )
{
Status = FmpOfflineWaitingTree(pResource);
if ( Status != ERROR_IO_PENDING)
{
FmpSignalGroupWaiters( pResource->Group );
}
}
}
else if (OldState == ClusterResourceOnlinePending)
{
FmpHandleResourceFailure(pResource);
FmpSignalGroupWaiters( pResource->Group );
}
break;
case ClusterResourceOnline:
if (OldState == ClusterResourceOnlinePending)
{
Status = FmpOnlineWaitingTree( pResource );
if (Status != ERROR_IO_PENDING) {
FmpSignalGroupWaiters( pResource->Group );
}
}
break;
case ClusterResourceOffline:
if ((OldState == ClusterResourceOfflinePending) ||
(OldState == ClusterResourceOffline))
{
Status = FmpOfflineWaitingTree(pResource);
if ( Status != ERROR_IO_PENDING)
{
FmpSignalGroupWaiters( pResource->Group );
}
}
break;
}
FnExit:
FmpReleaseLocalResourceLock( pResource );
if (bQuoChangeLockHeld) RELEASE_LOCK(gQuoChangeLock);
return;
}
VOID
FmpPostWorkItem(
IN CLUSTER_EVENT Event,
IN PVOID Context1,
IN ULONG_PTR Context2
)
/*++
Routine Description:
Posts a work item event to the FM work queue.
Arguments:
Event - The event to post.
Context1 - A pointer to some context. This context should be permanent
in memory - i.e. it should not be deallocated when this call
returns.
Context2 - A pointer to additional context. This context should be
permanent in memory - i.e. it should not be deallocated when this
call returns.
Returns:
None.
--*/
{
PWORK_ITEM workItem;
workItem = LocalAlloc(LMEM_FIXED, sizeof(WORK_ITEM));
if ( workItem == NULL ) {
CsInconsistencyHalt(ERROR_NOT_ENOUGH_MEMORY);
} else {
workItem->Event = Event;
workItem->Context1 = Context1;
workItem->Context2 = Context2;
//
// Insert work item on queue and wake up the worker thread.
//
ClRtlInsertTailQueue(&FmpWorkQueue, &workItem->ListEntry);
}
} // FmpPostEvent
DWORD
FmpStartWorkerThread(
VOID
)
{
DWORD threadId;
DWORD Status;
//
// Start up our worker thread
//
ClRtlLogPrint(LOG_NOISE,"[FM] Starting worker thread...\n");
FmpWorkerThreadHandle = CreateThread(
NULL,
0,
FmpWorkerThread,
NULL,
0,
&threadId
);
if (FmpWorkerThreadHandle == NULL) {
ClRtlLogPrint(LOG_NOISE,
"[FM] Failed to start worker thread %1!u!\n",
GetLastError()
);
return(GetLastError());
}
return(ERROR_SUCCESS);
} // FmpStartWorkerThread
BOOL
FmpAddNodeGroupCallback(
IN PVOID Context1,
IN PVOID Context2,
IN PVOID Object,
IN LPCWSTR Name
)
/*++
Routine Description:
Enumeration callback for each group in the system when a new
node is added to the cluster.
The algorithm used here is to enumerate all the resources in
this group. For each resource in the group that does not
have an explicit "PreferredOwners" setting in the registry,
the node is added as a PossibleNode. Finally, if the node
was added as a possiblenode for each resource in the group,
the node is added to the end of the preferredowners list for
the group.
Arguments:
Context1 - Supplies the PNM_NODE of the new node.
Context2 - Not used.
Object - Supplies the group object.
Name - Supplies the name of the group object.
Return Value:
TRUE
--*/
{
PFM_RESOURCE Resource;
PFM_GROUP Group;
PNM_NODE Node;
HDMKEY hKey;
DWORD Status;
PPREFERRED_ENTRY preferredEntry;
PRESOURCE_ENUM ResourceEnum;
DWORD i;
Group = (PFM_GROUP)Object;
Node = (PNM_NODE)Context1;
FmpAcquireLocalGroupLock( Group );
Status = FmpGetResourceList( &ResourceEnum,
Group );
FmpReleaseLocalGroupLock( Group );
if ( Status != ERROR_SUCCESS ) {
ClRtlLogPrint(LOG_NOISE,
"[FM] AddNodeGroup, failed to get resource list for group %1!ws!, status %2!u!.\n",
Name,
Status );
return(TRUE);
}
//
// First fix up the resource info.
//
for ( i = 0; i < ResourceEnum->EntryCount; i++ ) {
Resource = OmReferenceObjectById( ObjectTypeResource,
ResourceEnum->Entry[i].Id );
if ( Resource != NULL ) {
FmpAcquireLocalResourceLock( Resource );
//ss: we need to hold the resource lock as well
//since that is the one that the update (FmpUpdateChangeResourceNode)
// to remove and add nodes obtains and we must synchronize with it
FmpAcquireResourceLock();
Status = FmpFixupResourceInfo( Resource );
FmpReleaseResourceLock();
FmpReleaseLocalResourceLock( Resource );
if ( Status == ERROR_SUCCESS ) {
FmpRmResourceControl( Resource,
CLUSCTL_RESOURCE_INSTALL_NODE,
(PUCHAR)OmObjectName(Node),
((lstrlenW(OmObjectName(Node)) + 1) * sizeof(WCHAR)),
NULL,
0,
NULL,
NULL );
// Ignore status return
ClusterEvent( CLUSTER_EVENT_RESOURCE_PROPERTY_CHANGE,
Resource );
} else {
ClRtlLogPrint(LOG_UNUSUAL,
"[FM] AddNodeGroup, failed to fixup info for resource %1!ws! when node was added!\n",
OmObjectName( Resource ) );
}
OmDereferenceObject( Resource );
} else {
ClRtlLogPrint(LOG_NOISE,
"[FM] AddNodeGroup, failed to find resource %1!ws! in group %2!ws!.\n",
ResourceEnum->Entry[i].Id,
Name );
}
}
FmpDeleteResourceEnum( ResourceEnum );
//
// Now fix up the group information.
//
FmpAcquireLocalGroupLock( Group );
Status = FmpFixupGroupInfo( Group );
FmpReleaseLocalGroupLock( Group );
if ( Status == ERROR_SUCCESS ) {
ClusterEvent( CLUSTER_EVENT_GROUP_PROPERTY_CHANGE, Group );
} else {
ClRtlLogPrint(LOG_UNUSUAL,
"[FM] AddNodeGroup, failed to fixup info for group %1!ws! when node was added, status %2!u!.\n",
OmObjectName( Group ),
Status );
}
return(TRUE);
} // FmpAddNodeGroupCallback
Reference in a new issue Copy permalink