Source code of Windows XP (NT5)
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.
 
 
 
 
 
 

4876 lines
131 KiB

#include "precomp.h"
//
// MGC.CPP
// MCS Glue Layer, Legacy from simultaneous R.11 and T.120 support
//
// Copyright(c) Microsoft 1997-
//
#define MLZ_FILE_ZONE ZONE_NET
//
//
// CONSTANT DATA
//
// These arrays map between MCAT and DC-Grouwpare constants. They are not
// in separate data file since only referenced from this source file.
//
//
UINT McsErrToNetErr (UINT rcMCS);
const UINT c_RetCodeMap1[] =
{
0,
NET_RC_MGC_NOT_SUPPORTED,
NET_RC_MGC_NOT_INITIALIZED,
NET_RC_MGC_ALREADY_INITIALIZED,
NET_RC_MGC_INIT_FAIL,
NET_RC_MGC_INVALID_REMOTE_ADDRESS,
NET_RC_NO_MEMORY,
NET_RC_MGC_CALL_FAILED,
NET_RC_MGC_NOT_SUPPORTED,
NET_RC_MGC_NOT_SUPPORTED,
NET_RC_MGC_NOT_SUPPORTED, // security failed
};
const UINT c_RetCodeMap2[] =
{
NET_RC_MGC_DOMAIN_IN_USE,
NET_RC_MGC_INVALID_DOMAIN,
NET_RC_MGC_NOT_ATTACHED,
NET_RC_MGC_INVALID_USER_HANDLE,
NET_RC_MGC_TOO_MUCH_IN_USE,
NET_RC_MGC_INVALID_CONN_HANDLE,
NET_RC_MGC_INVALID_UP_DOWN_PARM,
NET_RC_MGC_NOT_SUPPORTED,
NET_RC_MGC_TOO_MUCH_IN_USE
};
#define MG_NUM_OF_MCS_RESULTS 15
#define MG_INVALID_MCS_RESULT MG_NUM_OF_MCS_RESULTS
NET_RESULT c_ResultMap[MG_NUM_OF_MCS_RESULTS+1] =
{
NET_RESULT_OK,
NET_RESULT_NOK,
NET_RESULT_NOK,
NET_RESULT_CHANNEL_UNAVAILABLE,
NET_RESULT_DOMAIN_UNAVAILABLE,
NET_RESULT_NOK,
NET_RESULT_REJECTED,
NET_RESULT_NOK,
NET_RESULT_NOK,
NET_RESULT_TOKEN_ALREADY_GRABBED,
NET_RESULT_TOKEN_NOT_OWNED,
NET_RESULT_NOK,
NET_RESULT_NOK,
NET_RESULT_NOT_SPECIFIED,
NET_RESULT_USER_REJECTED,
NET_RESULT_UNKNOWN
};
//
// MG_Register()
//
BOOL MG_Register
(
MGTASK task,
PMG_CLIENT * ppmgClient,
PUT_CLIENT putTask
)
{
PMG_CLIENT pmgClient = NULL;
CMTASK cmTask;
BOOL rc = FALSE;
DebugEntry(MG_Register);
UT_Lock(UTLOCK_T120);
ASSERT(task >= MGTASK_FIRST);
ASSERT(task < MGTASK_MAX);
//
// Check the putTask passed in:
//
ValidateUTClient(putTask);
//
// Does this already exist?
//
if (g_amgClients[task].putTask != NULL)
{
ERROR_OUT(("MG task %d already exists", task));
DC_QUIT;
}
pmgClient = &(g_amgClients[task]);
ZeroMemory(pmgClient, sizeof(MG_CLIENT));
pmgClient->putTask = putTask;
//
// Register an exit procedure
//
UT_RegisterExit(putTask, MGExitProc, pmgClient);
pmgClient->exitProcReg = TRUE;
//
// We register a high priority event handler (join by key handler) to
// intercept various events which are generated as part of the join by
// key processing. We register it now, before the call to
// MG_ChannelJoin below, to prevent events which we cant stop from
// going to the client if UT_RegisterEvent fails. This high priority
// handler also looks after our internal scheduling of pending
// requests.
//
UT_RegisterEvent(putTask, MGEventHandler, pmgClient, UT_PRIORITY_OBMAN);
pmgClient->eventProcReg = TRUE;
//
// Register our hidden event handler for the client (the parameter to
// be passed to the event handler is the pointer to the client CB):
//
UT_RegisterEvent(putTask, MGLongStopHandler, pmgClient, UT_PRIORITY_NETWORK);
pmgClient->lowEventProcReg = TRUE;
//
// Register as a call manager secondary.
//
switch (task)
{
case MGTASK_OM:
cmTask = CMTASK_OM;
break;
case MGTASK_DCS:
cmTask = CMTASK_DCS;
break;
default:
ASSERT(FALSE);
}
if (!CMS_Register(putTask, cmTask, &(pmgClient->pcmClient)))
{
ERROR_OUT(("CMS_Register failed"));
DC_QUIT;
}
rc = TRUE;
DC_EXIT_POINT:
*ppmgClient = pmgClient;
UT_Unlock(UTLOCK_T120);
DebugExitBOOL(MG_Register, rc);
return(rc);
}
//
// MG_Deregister(...)
//
void MG_Deregister(PMG_CLIENT * ppmgClient)
{
PMG_CLIENT pmgClient;
DebugEntry(MG_Deregister);
UT_Lock(UTLOCK_T120);
ASSERT(ppmgClient);
pmgClient = *ppmgClient;
ValidateMGClient(pmgClient);
MGExitProc(pmgClient);
//
// Dereg CMS handler. In abnormal situations, the CMS exit proc will
// clean it up for us.
//
if (pmgClient->pcmClient)
{
CMS_Deregister(&pmgClient->pcmClient);
}
*ppmgClient = NULL;
UT_Unlock(UTLOCK_T120);
DebugExitVOID(MG_Deregister);
}
//
// MGExitProc()
//
void CALLBACK MGExitProc(LPVOID uData)
{
PMG_CLIENT pmgClient = (PMG_CLIENT)uData;
PMG_BUFFER pmgBuffer;
DebugEntry(MGExitProc);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
//
// If the client has attached, detach it
//
if (pmgClient->userAttached)
{
MG_Detach(pmgClient);
}
//
// Free all buffers the client may be using:
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->buffers), FIELD_OFFSET(MG_BUFFER, clientChain));
while (pmgBuffer != NULL)
{
ValidateMGBuffer(pmgBuffer);
//
// This implicitly frees any user memory or MCS memory associated
// with the buffer CB.
//
MGFreeBuffer(pmgClient, &pmgBuffer);
//
// MGFreeBuffer removed this CB from the list, so we get the first
// one in what's left of the list - if the list is now empty, this
// will give us NULL and we will break out of the while loop:
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->buffers), FIELD_OFFSET(MG_BUFFER, clientChain));
}
//
// Deregister our event handler and exit procedure:
//
if (pmgClient->exitProcReg)
{
UT_DeregisterExit(pmgClient->putTask, MGExitProc, pmgClient);
pmgClient->exitProcReg = FALSE;
}
if (pmgClient->lowEventProcReg)
{
UT_DeregisterEvent(pmgClient->putTask, MGLongStopHandler, pmgClient);
pmgClient->lowEventProcReg = FALSE;
}
if (pmgClient->eventProcReg)
{
UT_DeregisterEvent(pmgClient->putTask, MGEventHandler, pmgClient);
pmgClient->eventProcReg = FALSE;
}
//
// We should only ever be asked to free a client CB which has had all
// of its child resources already freed, so do a quick sanity check:
//
ASSERT(pmgClient->buffers.next == 0);
//
// Set the putTask to NULL; that's how we know if a client is in use or
// not.
//
pmgClient->putTask = NULL;
UT_Unlock(UTLOCK_T120);
DebugExitVOID(MGExitProc);
}
//
// MG_Attach(...)
//
UINT MG_Attach
(
PMG_CLIENT pmgClient,
UINT callID,
PNET_FLOW_CONTROL pFlowControl
)
{
UINT rc = 0;
DebugEntry(MG_Attach);
UT_Lock(UTLOCK_T120);
ValidateCMP(g_pcmPrimary);
ValidateMGClient(pmgClient);
if (!g_pcmPrimary->callID)
{
//
// We aren't in a call yet/anymore.
//
WARNING_OUT(("MG_Attach failing; not in T.120 call"));
rc = NET_RC_MGC_NOT_CONNECTED;
DC_QUIT;
}
ASSERT(callID == g_pcmPrimary->callID);
ASSERT(!pmgClient->userAttached);
pmgClient->userIDMCS = NET_UNUSED_IDMCS;
ZeroMemory(&pmgClient->flo, sizeof(FLO_STATIC_DATA));
pmgClient->userAttached = TRUE;
//
// Call through to the underlying MCS layer (normally, we need our
// callbacks to happen with a task switch but since this is Windows it
// doesn't really matter anyway):
//
rc = MCS_AttachRequest(&(pmgClient->m_piMCSSap),
(DomainSelector) &g_pcmPrimary->callID,
sizeof(g_pcmPrimary->callID),
(MCSCallBack) MGCallback,
(void *) pmgClient,
ATTACHMENT_DISCONNECT_IN_DATA_LOSS);
if (rc != 0)
{
WARNING_OUT(("MCSAttachUserRequest failed with error %x", rc));
MGDetach(pmgClient);
rc = McsErrToNetErr(rc);
DC_QUIT;
}
if (++g_mgAttachCount == 1)
{
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
MG_TIMER_PERIOD,
NET_MG_WATCHDOG,
0, 0);
}
ASSERT(g_mgAttachCount <= MGTASK_MAX);
//
// It is assumed that the client will use the same latencies for every
// attachment, so we keep them at the client level.
//
pmgClient->flowControl = *pFlowControl;
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_Attach, rc);
return(rc);
}
//
// MG_Detach(...)
//
void MG_Detach
(
PMG_CLIENT pmgClient
)
{
DebugEntry(MG_Detach);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_Detach: client %x not attached", pmgClient));
DC_QUIT;
}
//
// Call FLO_UserTerm to ensure that flow control is stopped on all the
// channels that have been flow controlled on our behalf.
//
FLO_UserTerm(pmgClient);
//
// Clear out the buffers, variabls.
//
MGDetach(pmgClient);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitVOID(MG_Detach);
}
//
// MG_ChannelJoin(...)
//
UINT MG_ChannelJoin
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID * pCorrelator,
NET_CHANNEL_ID channel
)
{
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MG_ChannelJoin);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_ChannelJoin: client %x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// MCAT may bounce this request, so we must queue the request
//
rc = MGNewBuffer(pmgClient, MG_RQ_CHANNEL_JOIN, &pmgBuffer);
if (rc != 0)
{
DC_QUIT;
}
MGNewCorrelator(pmgClient, pCorrelator);
pmgBuffer->work = *pCorrelator;
pmgBuffer->channelId = (ChannelID)channel;
TRACE_OUT(("Inserting join message 0x%08x into pending chain", pmgBuffer));
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_ChannelJoin, rc);
return(rc);
}
//
// MG_ChannelJoinByKey(...)
//
UINT MG_ChannelJoinByKey
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID * pCorrelator,
WORD channelKey
)
{
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MG_ChannelJoinByKey);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_ChannelJoinByKey: client %x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// MCAT may bounce this request, so we must queue the request
//
rc = MGNewBuffer(pmgClient, MG_RQ_CHANNEL_JOIN_BY_KEY, &pmgBuffer);
if (rc != 0)
{
DC_QUIT;
}
//
// Store the various pieces of information in the joinByKeyInfo
// structure of the client CB
//
MGNewCorrelator(pmgClient, pCorrelator);
pmgBuffer->work = *pCorrelator;
pmgBuffer->channelKey = (ChannelID)channelKey;
pmgBuffer->channelId = 0;
TRACE_OUT(("Inserting join message 0x%08x into pending chain", pmgBuffer));
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_ChannelJoinByKey, rc);
return(rc);
}
//
// MG_ChannelLeave(...)
//
void MG_ChannelLeave
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID channel
)
{
PMG_BUFFER pmgBuffer;
DebugEntry(MG_ChannelLeave);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_ChannelLeave: client %x not attached", pmgClient));
DC_QUIT;
}
//
// MCAT may bounce this request, so instead of processing it straight
// away, we put it on the user's request queue and kick off a process
// queue loop: This is a request CB, but we don't need any data buffer
//
if (MGNewBuffer(pmgClient, MG_RQ_CHANNEL_LEAVE, &pmgBuffer) != 0)
{
DC_QUIT;
}
//
// Fill in the specific data fields in the request CB:
//
pmgBuffer->channelId = (ChannelID)channel;
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitVOID(MG_ChannelLeave);
}
//
// MG_SendData(...)
//
UINT MG_SendData
(
PMG_CLIENT pmgClient,
NET_PRIORITY priority,
NET_CHANNEL_ID channel,
UINT length,
void ** ppData
)
{
PMG_BUFFER pmgBuffer;
UINT numControlBlocks;
UINT i;
UINT rc;
DebugEntry(MG_SendData);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_SendData: client %x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// Check for a packet greater than the permitted size
// It must not cause the length to wrap into the flow flag
//
ASSERT(TSHR_MAX_SEND_PKT + sizeof(TSHR_NET_PKT_HEADER) < TSHR_PKT_FLOW);
ASSERT(length <= TSHR_MAX_SEND_PKT);
//
// Ensure we have a priority which is valid for our use of MCS.
//
priority = (NET_PRIORITY)(MG_VALID_PRIORITY(priority));
if (pmgClient->userIDMCS == NET_UNUSED_IDMCS)
{
//
// We are not yet attached, so don't try to send data.
//
ERROR_OUT(("Sending data prior to attach indication"));
rc = NET_RC_INVALID_STATE;
DC_QUIT;
}
//
// The <ppData> parameter points to a data buffer pointer. This buffer
// pointer should point to a buffer which the client acquired using
// MG_GetBuffer. MG_GetBuffer should have added a buffer CB to the
// client's buffer list containing the same pointer. Note that if the
// NET_SEND_ALL_PRIORITIES flag is set then there will be four buffers
// in the client's buffer list containing the same pointer.
//
// So, we search through the client's buffer list looking for a match
// on the data buffer pointer. Move to the first position in the list.
//
COM_BasedListFind(LIST_FIND_FROM_FIRST, &(pmgClient->buffers),
(void**)&pmgBuffer, FIELD_OFFSET(MG_BUFFER, clientChain),
FIELD_OFFSET(MG_BUFFER, pDataBuffer), (DWORD_PTR)*ppData,
FIELD_SIZE(MG_BUFFER, pDataBuffer));
ValidateMGBuffer(pmgBuffer);
//
// Check the NET_SEND_ALL_PRIORITIES flag to see if it is set
//
if (pmgBuffer->priority & NET_SEND_ALL_PRIORITIES)
{
//
// Check that the priority and channel has not changed. Changing
// the priority between calling MG_GetBuffer and calling
// MG_SendData is not allowed.
//
ASSERT(pmgBuffer->channelId == channel);
ASSERT(priority & NET_SEND_ALL_PRIORITIES);
//
// The flag is set so there should be multiple control buffers
// waiting to be sent.
//
numControlBlocks = MG_NUM_PRIORITIES;
}
else
{
//
// Check that the priority and channel has not changed.
//
ASSERT(pmgBuffer->channelId == channel);
ASSERT(pmgBuffer->priority == priority);
//
// The flag is not set so there should be only one control buffer
// waiting.
//
numControlBlocks = 1;
}
//
// Now send the control blocks
//
for (i = 0; i < numControlBlocks; i++)
{
ValidateMGBuffer(pmgBuffer);
//
// Clear the NET_SEND_ALL_PRIORITIES flag.
//
pmgBuffer->priority &= ~NET_SEND_ALL_PRIORITIES;
//
// Set up the packet length for the send (this may be different
// from the length in the buffer header since the app may not have
// used all the buffer).
//
ASSERT(length + sizeof(TSHR_NET_PKT_HEADER) <= pmgBuffer->length);
pmgBuffer->pPktHeader->header.pktLength = (TSHR_UINT16)(length + sizeof(TSHR_NET_PKT_HEADER));
//
// If the length has changed then tell FC about it.
//
if ((length + sizeof(MG_INT_PKT_HEADER)) < pmgBuffer->length)
{
FLO_ReallocSend(pmgClient, pmgBuffer->pStr,
pmgBuffer->length - (length + sizeof(MG_INT_PKT_HEADER)));
}
TRACE_OUT(("Inserting send 0x%08x into pend chain, pri %u, chan 0x%08x",
pmgBuffer, pmgBuffer->priority, pmgBuffer->channelId));
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
//
// If there is one or more control block left to find then search
// the client's buffer list for it.
//
if ((numControlBlocks - (i + 1)) > 0)
{
COM_BasedListFind(LIST_FIND_FROM_NEXT, &(pmgClient->buffers),
(void**)&pmgBuffer, FIELD_OFFSET(MG_BUFFER, clientChain),
FIELD_OFFSET(MG_BUFFER, pDataBuffer),
(DWORD_PTR)*ppData, FIELD_SIZE(MG_BUFFER, pDataBuffer));
}
}
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
//
// Everything went OK - set the ppData pointer to NULL to prevent
// the caller from accessing the memory.
//
*ppData = NULL;
rc = 0;
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_SendData, rc);
return(rc);
}
//
// MG_TokenGrab(...)
//
UINT MG_TokenGrab
(
PMG_CLIENT pmgClient,
NET_TOKEN_ID tokenID
)
{
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MG_TokenGrab);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_TokenGrab: client 0x%08x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// MCAT may bounce this request, so instead of processing it straight
// away, we put it on the user's request queue and kick off a process
// queue loop:
//
rc = MGNewBuffer(pmgClient, MG_RQ_TOKEN_GRAB, &pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewBuffer failed in MG_TokenGrab"));
DC_QUIT;
}
pmgBuffer->channelId = (ChannelID)tokenID;
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_TokenGrab, rc);
return(rc);
}
//
// MG_TokenInhibit(...)
//
UINT MG_TokenInhibit
(
PMG_CLIENT pmgClient,
NET_TOKEN_ID tokenID
)
{
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MG_TokenInhibit);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_TokenInhibit: client 0x%08x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// MCAT may bounce this request, so instead of processing it straight
// away, we put it on the user's request queue and kick off a process
// queue loop:
//
rc = MGNewBuffer(pmgClient, MG_RQ_TOKEN_INHIBIT, &pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewBuffer failed in MG_TokenInhibit"));
DC_QUIT;
}
pmgBuffer->channelId = (ChannelID)tokenID;
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_TokenInhibit, rc);
return(rc);
}
//
// MG_GetBuffer(...)
//
UINT MG_GetBuffer
(
PMG_CLIENT pmgClient,
UINT length,
NET_PRIORITY priority,
NET_CHANNEL_ID channel,
void ** ppData
)
{
PMG_BUFFER pmgBuffer;
UINT rc;
DebugEntry(MG_GetBuffer);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_GetBuffer: client 0x%08x not attached", pmgClient));
rc = NET_RC_MGC_INVALID_USER_HANDLE;
DC_QUIT;
}
//
// Ensure we have a priority which is valid for our use of MCS.
//
priority = (NET_PRIORITY)(MG_VALID_PRIORITY(priority));
//
// Obtain a buffer and store the info in a buffer CB hung off the
// client's list:
//
rc = MGNewTxBuffer(pmgClient, priority, channel, length,
&pmgBuffer);
if (rc != 0)
{
DC_QUIT;
}
//
// We always return a pointer to the data buffer to an application.
// The MG packet header is only used when giving data to MCS or
// receiving data from MCS.
//
*ppData = pmgBuffer->pDataBuffer;
DC_EXIT_POINT:
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MG_GetBuffer, rc);
return(rc);
}
//
// MG_FreeBuffer(...)
//
void MG_FreeBuffer
(
PMG_CLIENT pmgClient,
void ** ppData
)
{
PMG_BUFFER pmgBuffer;
DebugEntry(MG_FreeBuffer);
UT_Lock(UTLOCK_T120);
ValidateMGClient(pmgClient);
//
// Find the buffer CB associated with the buffer - an application
// always uses a pointer to the data buffer rather than the packet
// header.
//
COM_BasedListFind(LIST_FIND_FROM_FIRST, &(pmgClient->buffers),
(void**)&pmgBuffer, FIELD_OFFSET(MG_BUFFER, clientChain),
FIELD_OFFSET(MG_BUFFER, pDataBuffer), (DWORD_PTR)*ppData,
FIELD_SIZE(MG_BUFFER, pDataBuffer));
ValidateMGBuffer(pmgBuffer);
//
// If the app is freeing a send buffer (e.g. because it decided not to
// send it) then inform flow control:
//
if (pmgBuffer->type == MG_TX_BUFFER)
{
FLO_ReallocSend(pmgClient,
pmgBuffer->pStr,
pmgBuffer->length);
}
//
// Now free the buffer CB and all associated data:
//
MGFreeBuffer(pmgClient, &pmgBuffer);
//
// Reset the client's pointer:
//
*ppData = NULL;
UT_Unlock(UTLOCK_T120);
DebugExitVOID(MG_FreeBuffer);
}
//
// MG_FlowControlStart
//
void MG_FlowControlStart
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID channel,
NET_PRIORITY priority,
UINT backlog,
UINT maxBytesOutstanding
)
{
DebugEntry(MG_FlowControlStart);
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MG_FlowControlStart: client 0x%08x not attached", pmgClient));
DC_QUIT;
}
//
// Ensure we have a priority which is valid for our use of MCS.
//
priority = (NET_PRIORITY)(MG_VALID_PRIORITY(priority));
FLO_StartControl(pmgClient,
channel,
priority,
backlog,
maxBytesOutstanding);
DC_EXIT_POINT:
DebugExitVOID(MG_FlowControlStart);
}
//
// MGLongStopHandler(...)
//
BOOL CALLBACK MGLongStopHandler
(
LPVOID pData,
UINT event,
UINT_PTR UNUSEDparam1,
UINT_PTR param2
)
{
PMG_CLIENT pmgClient;
BOOL processed = FALSE;
DebugEntry(MGLongStopHandler);
pmgClient = (PMG_CLIENT)pData;
ValidateMGClient(pmgClient);
if (event == NET_EVENT_CHANNEL_JOIN)
{
WARNING_OUT(("Failed to process NET_EVENT_CHANNEL_JOIN; freeing buffer 0x%08x",
param2));
MG_FreeBuffer(pmgClient, (void **)&param2);
processed = TRUE;
}
else if (event == NET_FLOW)
{
WARNING_OUT(("Failed to process NET_FLOW; freeing buffer 0x%08x",
param2));
processed = TRUE;
}
DebugExitBOOL(MGLongStopHandler, processed);
return(processed);
}
//
// MGEventHandler(...)
//
BOOL CALLBACK MGEventHandler
(
LPVOID pData,
UINT event,
UINT_PTR param1,
UINT_PTR param2
)
{
PMG_CLIENT pmgClient;
PNET_JOIN_CNF_EVENT pNetJoinCnf = NULL;
BOOL processed = TRUE;
PMG_BUFFER pmgBuffer;
BOOL joinComplete = FALSE;
UINT result = NET_RESULT_USER_REJECTED;
DebugEntry(MGEventHandler);
pmgClient = (PMG_CLIENT)pData;
ValidateMGClient(pmgClient);
switch (event)
{
case NET_EVENT_CHANNEL_JOIN:
{
//
// If there are no join requests queued off the client CB then
// we have nothing more to do. The only NET events we are
// interested in are NET_EV_JOIN_CONFIRM events - pass any others
// on.
//
if (pmgClient->joinChain.next == 0)
{
//
// Pass the event on...
//
processed = FALSE;
DC_QUIT;
}
//
// We must be careful not to process a completed channel join
// which we intend to go to the client. The correlator is only
// filled in on completed events and is always non-zero.
//
pNetJoinCnf = (PNET_JOIN_CNF_EVENT)param2;
if (pNetJoinCnf->correlator != 0)
{
//
// Pass the event on...
//
processed = FALSE;
DC_QUIT;
}
//
// There is only ever one join request outstanding per client,
// so the join confirm is for the first join request in the
// list.
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->joinChain),
FIELD_OFFSET(MG_BUFFER, pendChain));
ValidateMGBuffer(pmgBuffer);
//
// We will post a join confirm to the application. Set up the
// parameters which are needed.
//
result = pNetJoinCnf->result;
//
// Assume for now that we have completed the pending join
// request.
//
joinComplete = TRUE;
//
// If the result is a failure, we've finished
//
if (result != NET_RESULT_OK)
{
WARNING_OUT(("Failed to join channel 0x%08x, result %u",
pmgBuffer->channelId,
pNetJoinCnf->result));
DC_QUIT;
}
//
// The join request was successful. There are three different
// scenarios for issuing a join request...
//
// (a) A regular channel join.
// (b) Stage 1 of a channel join by key (get MCS to assign a
// channel number, which we will try to register).
// (c) Stage 2 of a channel join by key (join the registered
// channel).
//
if (pmgBuffer->type == MG_RQ_CHANNEL_JOIN)
{
//
// This is the completion of a regular channel join. Copy
// the channel Id from the join confirm to the bufferCB
// (the join request may have been for channel 0).
//
pmgBuffer->channelId = (ChannelID)pNetJoinCnf->channel;
TRACE_OUT(("Channel join complete, channel 0x%08x",
pmgBuffer->channelId));
DC_QUIT;
}
//
// This is channel join by key
//
if (pmgBuffer->channelId != 0)
{
//
// This is the completion of a channel join by key.
//
TRACE_OUT(("Channel join by key complete, channel 0x%08x, key %d",
pmgBuffer->channelId,
pmgBuffer->channelKey));
DC_QUIT;
}
//
// This is Stage 1 of a channel join by key. Fill in the
// channel Id which MCS has assigned us into the bufferCB,
// otherwise we'll lose track of the channel Id which we're
// registering.
//
pmgBuffer->channelId = (ChannelID)pNetJoinCnf->channel;
//
// This must be completion of stage 1 of a join by key. We now
// have to register the channel Id.
//
TRACE_OUT(("Registering channel 0x%08x, key %d",
pmgBuffer->channelId,
pmgBuffer->channelKey));
if (!CMS_ChannelRegister(pmgClient->pcmClient,
pmgBuffer->channelKey,
pmgBuffer->channelId))
{
WARNING_OUT(("Failed to register channel, "
"channel 0x%08x, key %d, result %u",
pmgBuffer->channelId,
pmgBuffer->channelKey,
param1));
//
// This causes us to post an error notification
//
result = NET_RESULT_USER_REJECTED;
DC_QUIT;
}
TRACE_OUT(("Waiting for CMS_CHANNEL_REGISTER_CONFIRM"));
//
// We're now waiting for a CMS_CHANNEL_REGISTER_CONFIRM, so we
// haven't finished processing the join request
//
joinComplete = FALSE;
break;
}
case CMS_CHANNEL_REGISTER_CONFIRM:
{
//
// If there are no join requests queued off the client CB then
// we have nothing more to do.
//
if (pmgClient->joinChain.next == 0)
{
processed = FALSE;
DC_QUIT;
}
TRACE_OUT(("CMS_CHANNEL_REGISTER rcvd, result %u, channel %u",
param1, param2));
//
// Assume for now that we have completed the pending join
// request.
//
joinComplete = TRUE;
//
// There is only ever one join request outstanding per client,
// so the channel register confirm is for the first join
// request in the list.
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->joinChain),
FIELD_OFFSET(MG_BUFFER, pendChain));
ValidateMGBuffer(pmgBuffer);
//
// Param1 contains the result, LOWORD(param2) contains the
// channel number of the registered channel (NOT necessarily
// the same as the channel we tried to register).
//
if (!param1)
{
WARNING_OUT(("Failed to register channel, "
"channel 0x%08x, key %d, result %u",
pmgBuffer->channelId,
pmgBuffer->channelKey,
param1));
result = NET_RESULT_USER_REJECTED;
DC_QUIT;
}
//
// If the channel number returned in the confirm event is the
// same as the channel number which we tried to register, then
// we have finished. Otherwise we have to leave the channel we
// tried to register and join the channel returned instead.
//
if (LOWORD(param2) == pmgBuffer->channelId)
{
TRACE_OUT(("Channel join by key complete, "
"channel 0x%08x, key %d",
pmgBuffer->channelId,
pmgBuffer->channelKey));
result = NET_RESULT_OK;
DC_QUIT;
}
MG_ChannelLeave(pmgClient, pmgBuffer->channelId);
pmgBuffer->channelId = (ChannelID)LOWORD(param2);
//
// Now we simply requeue the request onto the pending execution
// chain, but now with a set channel id to join
//
TRACE_OUT(("Inserting 0x%08x into pending chain",pmgBuffer));
COM_BasedListRemove(&(pmgBuffer->pendChain));
COM_BasedListInsertBefore(&(pmgClient->pendChain),
&(pmgBuffer->pendChain));
//
// We are now waiting for a join confirm (we've not finished
// yet !). However, we've requeued the bufferCB, so we can now
// process another join request (or the one we've requeued if
// its the only one).
//
joinComplete = FALSE;
pmgClient->joinPending = FALSE;
MGProcessPendingQueue(pmgClient);
break;
}
case NET_MG_SCHEDULE:
{
MGProcessPendingQueue(pmgClient);
break;
}
case NET_MG_WATCHDOG:
{
MGProcessDomainWatchdog(pmgClient);
break;
}
default:
{
//
// Don't do anything - we want to pass this event on.
//
processed = FALSE;
break;
}
}
DC_EXIT_POINT:
if (processed && pNetJoinCnf)
{
//
// Call MG_FreeBuffer to free up the event memory (we know that
// MG_FreeBuffer doesn't use the hUser so we pass in zero):
//
MG_FreeBuffer(pmgClient, (void **)&pNetJoinCnf);
}
if (joinComplete)
{
//
// We have either completed the channel join, or failed -
// either way we have finished processing the join request.
//
// We have to:
// - post a NET_EVENT_CHANNEL_JOIN event to the client
// - free up the bufferCB
// - reset the client's joinPending state
//
MGPostJoinConfirm(pmgClient,
(NET_RESULT)result,
pmgBuffer->channelId,
(NET_CHANNEL_ID)pmgBuffer->work);
MGFreeBuffer(pmgClient, &pmgBuffer);
pmgClient->joinPending = FALSE;
}
DebugExitBOOL(MGEventHandler, processed);
return(processed);
}
//
// MGCallback(...)
//
#ifdef _DEBUG
const char * c_szMCSMsgTbl[22] =
{
"MCS_CONNECT_PROVIDER_INDICATION", // 0
"MCS_CONNECT_PROVIDER_CONFIRM", // 1
"MCS_DISCONNECT_PROVIDER_INDICATION", // 2
"MCS_ATTACH_USER_CONFIRM", // 3
"MCS_DETACH_USER_INDICATION", // 4
"MCS_CHANNEL_JOIN_CONFIRM", // 5
"MCS_CHANNEL_LEAVE_INDICATION", // 6
"MCS_CHANNEL_CONVENE_CONFIRM", // 7
"MCS_CHANNEL_DISBAND_INDICATION", // 8
"MCS_CHANNEL_ADMIT_INDICATION", // 9
"MCS_CHANNEL_EXPEL_INDICATION", // 10
"MCS_SEND_DATA_INDICATION", // 11
"MCS_UNIFORM_SEND_DATA_INDICATION", // 12
"MCS_TOKEN_GRAB_CONFIRM", // 13
"MCS_TOKEN_INHIBIT_CONFIRM", // 14
"MCS_TOKEN_GIVE_INDICATION", // 15
"MCS_TOKEN_GIVE_CONFIRM", // 16
"MCS_TOKEN_PLEASE_INDICATION", // 17
"MCS_TOKEN_RELEASE_CONFIRM", // 18
"MCS_TOKEN_TEST_CONFIRM", // 19
"MCS_TOKEN_RELEASE_INDICATION", // 20
"MCS_TRANSMIT_BUFFER_AVAILABLE_INDICATION", // 21
};
// MCS_MERGE_DOMAIN_INDICATION 200
// MCS_TRANSPORT_STATUS_INDICATION 101
char * DbgGetMCSMsgStr(unsigned short mcsMessageType)
{
if (mcsMessageType <= 21)
{
return (char *) c_szMCSMsgTbl[mcsMessageType];
}
#ifdef USE_MERGE_DOMAIN_CODE
else if (mcsMessageType == MCS_MERGE_DOMAIN_INDICATION)
{
return "MCS_MERGE_DOMAIN_INDICATION";
}
#endif // USE_MERGE_DOMAIN_CODE
else if (mcsMessageType == MCS_TRANSPORT_STATUS_INDICATION)
{
return "MCS_TRANSPORT_STATUS_INDICATION";
}
return "Unknown";
}
#endif // _DEBUG
void CALLBACK MGCallback
(
unsigned int mcsMessageType,
unsigned long eventData,
unsigned long pData
)
{
PMG_CLIENT pmgClient;
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MGCallback);
UT_Lock(UTLOCK_T120);
pmgClient = (PMG_CLIENT)pData;
ValidateMGClient(pmgClient);
if (!pmgClient->userAttached)
{
TRACE_OUT(("MGCallback: client 0x%08x not attached", pmgClient));
DC_QUIT;
}
ValidateCMP(g_pcmPrimary);
switch (mcsMessageType)
{
case MCS_UNIFORM_SEND_DATA_INDICATION:
case MCS_SEND_DATA_INDICATION:
{
//
// The processing for a SEND_DATA_INDICATION is complicated
// significantly by MCS segmenting packets, so we call
// MGHandleSendInd to do all the work , then quit out of the
// function rather than special casing throughout.
//
rc = MGHandleSendInd(pmgClient, (PSendData)eventData);
DC_QUIT;
break;
}
case MCS_ATTACH_USER_CONFIRM:
{
NET_UID user;
NET_RESULT result;
user = LOWUSHORT(eventData);
result = TranslateResult(HIGHUSHORT(eventData));
//
// If the attach did not succeed, clean up:
//
if (HIGHUSHORT(eventData) != RESULT_SUCCESSFUL)
{
WARNING_OUT(("MG_Attach failed; cleaning up"));
MGDetach(pmgClient);
}
else
{
pmgClient->userIDMCS = user;
//
// Now initialize flow control for this user attachment
//
ZeroMemory(&(pmgClient->flo), sizeof(pmgClient->flo));
pmgClient->flo.callBack = MGFLOCallBack;
}
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_USER_ATTACH, MAKELONG(user, result),
g_pcmPrimary->callID);
break;
}
case MCS_DETACH_USER_INDICATION:
{
NET_UID user;
user = LOWUSHORT(eventData);
//
// If the detach is for the local user, then clean up
// the user CB:
//
if (user == pmgClient->userIDMCS)
{
//
// First terminate flow control
//
FLO_UserTerm(pmgClient);
MGDetach(pmgClient);
}
else
{
//
// Just remove the offending user from flow control
//
FLO_RemoveUser(pmgClient, user);
}
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_USER_DETACH, user, g_pcmPrimary->callID);
break;
}
case MCS_CHANNEL_JOIN_CONFIRM:
{
PNET_JOIN_CNF_EVENT pNetEvent;
UINT i;
//
// Allocate a buffer for the event
//
rc = MGNewDataBuffer(pmgClient, MG_EV_BUFFER,
sizeof(MG_INT_PKT_HEADER) + sizeof(NET_JOIN_CNF_EVENT), &pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewDataBuffer failed in MGCallback"));
DC_QUIT;
}
pNetEvent = (PNET_JOIN_CNF_EVENT)pmgBuffer->pDataBuffer;
//
// Fill in the call ID:
//
pNetEvent->callID = g_pcmPrimary->callID;
pNetEvent->channel = LOWUSHORT(eventData);
pNetEvent->result = TranslateResult(HIGHUSHORT(eventData));
//
// Now establish flow control for the newly joined channel
// Only control priorities that have a non-zero latency
// And remember to ignore our own user channel! And top priority.
//
if (HIGHUSHORT(eventData) == RESULT_SUCCESSFUL)
{
if (pNetEvent->channel != pmgClient->userIDMCS)
{
for (i = 0; i < NET_NUM_PRIORITIES; i++)
{
if ((i == MG_VALID_PRIORITY(i)) &&
(pmgClient->flowControl.latency[i] != 0))
{
FLO_StartControl(pmgClient, pNetEvent->channel,
i, pmgClient->flowControl.latency[i],
pmgClient->flowControl.streamSize[i]);
}
}
}
}
//
// OK, we've built the DCG event so now post it to our client:
//
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_CHANNEL_JOIN, 0, (UINT_PTR)pNetEvent);
pmgBuffer->eventPosted = TRUE;
break;
}
case MCS_CHANNEL_LEAVE_INDICATION:
{
NET_CHANNEL_ID channel;
channel = LOWUSHORT(eventData);
MGProcessEndFlow(pmgClient, channel);
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_CHANNEL_LEAVE, channel, g_pcmPrimary->callID);
break;
}
case MCS_TOKEN_GRAB_CONFIRM:
{
NET_RESULT result;
result = TranslateResult(HIGHUSHORT(eventData));
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_TOKEN_GRAB, result, g_pcmPrimary->callID);
break;
}
case MCS_TOKEN_INHIBIT_CONFIRM:
{
NET_RESULT result;
result = TranslateResult(HIGHUSHORT(eventData));
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_TOKEN_INHIBIT, result, g_pcmPrimary->callID);
break;
}
default:
break;
}
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_MG_SCHEDULE, 0, 0);
DC_EXIT_POINT:
if (rc != 0)
{
//
// We hit an error, but must return OK to MCS - otherwise it will
// keep sending us the callback forever!
//
WARNING_OUT(("MGCallback: Error 0x%08x processing MCS message %u",
rc, mcsMessageType));
}
UT_Unlock(UTLOCK_T120);
DebugExitDWORD(MGCallback, MCS_NO_ERROR);
}
//
// ProcessEndFlow(...)
//
void MGProcessEndFlow
(
PMG_CLIENT pmgClient,
ChannelID channel
)
{
UINT i;
DebugEntry(MGProcessEndFlow);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
//
// Terminate flow control for the newly left channel
//
if (channel != pmgClient->userIDMCS)
{
for (i = 0; i < NET_NUM_PRIORITIES; i++)
{
if ((i == MG_VALID_PRIORITY(i)) &&
(pmgClient->flowControl.latency[i] != 0))
{
TRACE_OUT(("Ending flow control on channel 0x%08x priority %u",
channel, i));
FLO_EndControl(pmgClient, channel, i);
}
}
}
DebugExitVOID(MGProcessEndFlow);
}
//
// MGHandleSendInd(...)
//
UINT MGHandleSendInd
(
PMG_CLIENT pmgClient,
PSendData pSendData
)
{
PMG_BUFFER pmgBuffer;
PNET_SEND_IND_EVENT pEvent;
NET_PRIORITY priority;
LPBYTE pData;
UINT cbData;
UINT rc = 0;
TSHR_NET_PKT_HEADER pktHeader;
DebugEntry(MGHandleSendInd);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
priority = (NET_PRIORITY)MG_VALID_PRIORITY(
(NET_PRIORITY)pSendData->data_priority);
pData = pSendData->user_data.value;
ASSERT(pData != NULL);
cbData = pSendData->user_data.length;
ASSERT(cbData > sizeof(TSHR_NET_PKT_HEADER));
TRACE_OUT(("MCS Data Indication: flags 0x%08x, size %u, first dword 0x%08x",
pSendData->segmentation, pSendData->user_data.length,
*((DWORD *)pData)));
ASSERT (pSendData->segmentation == (SEGMENTATION_BEGIN | SEGMENTATION_END));
TRACE_OUT(("Only segment: channel %u, priority %u, length %u",
pSendData->channel_id, pSendData->data_priority, cbData));
//
// Look at the header
//
memcpy(&pktHeader, pData, sizeof(TSHR_NET_PKT_HEADER));
//
// Trace out the MG header word
//
TRACE_OUT(("Got 1st MG segment (header=%X)", pktHeader.pktLength));
//
// First of all try for a flow control packet
//
if (pktHeader.pktLength & TSHR_PKT_FLOW)
{
TRACE_OUT(("Flow control packet"));
if (pktHeader.pktLength == TSHR_PKT_FLOW)
{
FLO_ReceivedPacket(pmgClient,
(PTSHR_FLO_CONTROL)(pData + sizeof(TSHR_NET_PKT_HEADER)));
}
else
{
WARNING_OUT(("Received obsolete throughput packet size 0x%04x", pktHeader.pktLength));
}
pmgClient->m_piMCSSap->FreeBuffer((PVoid) pData);
DC_QUIT;
}
//
// Allocate headers for the incoming buffer.
//
//
ASSERT((sizeof(NET_SEND_IND_EVENT) + pktHeader.pktLength) <= 0xFFFF);
ASSERT(pktHeader.pktLength == cbData);
rc = MGNewRxBuffer(pmgClient,
priority,
pSendData->channel_id,
pSendData->initiator,
&pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewRxBuffer of size %u failed",
sizeof(NET_SEND_IND_EVENT) + sizeof(MG_INT_PKT_HEADER)));
pmgClient->m_piMCSSap->FreeBuffer((PVoid) pData);
DC_QUIT;
}
pEvent = (PNET_SEND_IND_EVENT) pmgBuffer->pDataBuffer;
ValidateCMP(g_pcmPrimary);
pEvent->callID = g_pcmPrimary->callID;
pEvent->priority = priority;
pEvent->channel = pSendData->channel_id;
//
// Copy the length into the data buffer header.
//
pmgBuffer->pPktHeader->header = pktHeader;
//
// We want to skip past the packet header to the user data
//
pData += sizeof(TSHR_NET_PKT_HEADER);
cbData -= sizeof(TSHR_NET_PKT_HEADER);
//
// Set the pointer in the buffer header to point to the received data.
//
// pEvent->lengthOfData contains the number of bytes received in this
// event so far.
//
ASSERT(pData);
pEvent->data_ptr = pData;
pEvent->lengthOfData = cbData;
TRACE_OUT(("New RX pmgBuffer 0x%08x pDataBuffer 0x%08x",
pmgBuffer, pEvent));
//
// OK, we've got all the segments, so post it to our client:
//
UT_PostEvent(pmgClient->putTask, pmgClient->putTask, NO_DELAY,
NET_EVENT_DATA_RECEIVED, 0, (UINT_PTR)pEvent);
pmgBuffer->eventPosted = TRUE;
DC_EXIT_POINT:
DebugExitDWORD(MGHandleSendInd, rc);
return(rc);
}
//
// MGNewBuffer(...)
//
UINT MGNewBuffer
(
PMG_CLIENT pmgClient,
UINT bufferType,
PMG_BUFFER * ppmgBuffer
)
{
PMG_BUFFER pmgBuffer;
void * pBuffer = NULL;
UINT rc = 0;
DebugEntry(MGNewBuffer);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
pmgBuffer = new MG_BUFFER;
if (!pmgBuffer)
{
WARNING_OUT(("MGNewBuffer failed; out of memory"));
rc = NET_RC_NO_MEMORY;
DC_QUIT;
}
ZeroMemory(pmgBuffer, sizeof(*pmgBuffer));
SET_STAMP(pmgBuffer, MGBUFFER);
pmgBuffer->type = bufferType;
//
// Insert it at the head of this client's list of allocated buffers:
//
COM_BasedListInsertAfter(&(pmgClient->buffers), &(pmgBuffer->clientChain));
//
// return the pointer
//
*ppmgBuffer = pmgBuffer;
DC_EXIT_POINT:
DebugExitDWORD(MGNewBuffer, rc);
return(rc);
}
//
// MGNewDataBuffer(...)
//
UINT MGNewDataBuffer
(
PMG_CLIENT pmgClient,
UINT bufferType,
UINT bufferSize,
PMG_BUFFER * ppmgBuffer
)
{
void * pBuffer = NULL;
PMG_BUFFER pmgBuffer;
UINT rc = 0;
DebugEntry(MGNewDataBuffer);
//
// Buffers include an MG internal packet header that has a length field
// which we add to the start of all user data passed to/received from
// MCS. This is four byte aligned, and since the data buffer starts
// immediately after this, the data buffer will be aligned.
//
pBuffer = new BYTE[bufferSize];
if (!pBuffer)
{
WARNING_OUT(("MGNewDataBuffer allocation of size %u failed", bufferSize));
rc = NET_RC_NO_MEMORY;
DC_QUIT;
}
ZeroMemory(pBuffer, bufferSize);
//
// Now we allocate the buffer CB which we will use to track the use of
// the buffer.
//
rc = MGNewBuffer(pmgClient, bufferType, ppmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewBuffer failed"));
DC_QUIT;
}
//
// Initialise the buffer entry
//
pmgBuffer = *ppmgBuffer;
pmgBuffer->length = bufferSize;
pmgBuffer->pPktHeader = (PMG_INT_PKT_HEADER)pBuffer;
pmgBuffer->pDataBuffer = (LPBYTE)pBuffer + sizeof(MG_INT_PKT_HEADER);
//
// Initialize the use count of the data buffer
//
pmgBuffer->pPktHeader->useCount = 1;
DC_EXIT_POINT:
if (rc != 0)
{
//
// Cleanup:
//
if (pBuffer != NULL)
{
WARNING_OUT(("Freeing MG_BUFFER data 0x%08x; MGNewBuffer failed", pBuffer));
delete[] pBuffer;
}
}
DebugExitDWORD(MGNewDataBuffer, rc);
return(rc);
}
//
// MGNewTxBuffer(...)
//
UINT MGNewTxBuffer
(
PMG_CLIENT pmgClient,
NET_PRIORITY priority,
NET_CHANNEL_ID channel,
UINT bufferSize,
PMG_BUFFER * ppmgBuffer
)
{
int i;
UINT numPrioritiesToUse;
UINT rc = 0;
UINT nextPriority;
PMG_BUFFER pmgBufferArray[MG_NUM_PRIORITIES];
PFLO_STREAM_DATA pStr[MG_NUM_PRIORITIES];
NET_PRIORITY priorities[MG_NUM_PRIORITIES];
DebugEntry(MGNewTxBuffer);
ValidateMGClient(pmgClient);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Initialise the control buffer pointer array. The first member of
// this array is the normal buffer which is allocated regardless of the
// NET_SEND_ALL_PRIORITIES flag. The remaining members are used for
// duplicate control buffer pointers needed for sending data on all
// priorities.
//
ZeroMemory(pmgBufferArray, sizeof(pmgBufferArray));
ZeroMemory(pStr, sizeof(pStr));
//
// SFR6025: Check for the NET_SEND_ALL_PRIORITIES flag. This means
// that the data will be sent at all four priorities. If it
// is not set then we just need to send data at one priority.
// In either case we need to:
//
// Check with flow control that it is possible to send data on
// all channels
//
// Allocate an additional three control blocks which all point
// to the same data block and bump up the usage count.
//
//
// NOTE: Previously this function just checked with flow control for
// a single channel.
//
if (priority & NET_SEND_ALL_PRIORITIES)
{
numPrioritiesToUse = MG_NUM_PRIORITIES;
}
else
{
numPrioritiesToUse = 1;
}
//
// Disable the flag to prevent FLO_AllocSend being sent an invalid
// priority.
//
priority &= ~NET_SEND_ALL_PRIORITIES;
nextPriority = priority;
for (i = 0; i < (int) numPrioritiesToUse; i++)
{
//
// Check with flow control to ensure that send space is available.
// Start with the requested priority level and continue for the
// other priority levels.
//
priorities[i] = (NET_PRIORITY)nextPriority;
rc = FLO_AllocSend(pmgClient,
nextPriority,
channel,
bufferSize + sizeof(MG_INT_PKT_HEADER),
&(pStr[i]));
//
// If we have got back pressure then just return.
//
if (rc != 0)
{
TRACE_OUT(("Received back pressure"));
//
// Free any buffer space allocated by FLO_AllocSend.
//
for ( --i; i >= 0; i--)
{
FLO_ReallocSend(pmgClient,
pStr[i],
bufferSize + sizeof(MG_INT_PKT_HEADER));
}
DC_QUIT;
}
ValidateFLOStr(pStr[i]);
//
// Move on to the next priority level. There are MG_NUM_PRIORITY
// levels, numbered contiguously from MG_PRIORITY_HIGHEST. The
// first priority processed can be any level in the valid range so
// rather than simply add 1 to get to the next level, we need to
// cope with the wrap-around back to MG_PRIORITY_HIGHEST when we
// have just processed the last priority, ie MG_PRIORITY_HIGHEST +
// MG_NUM_PRIORITIES - 1. This is achieved by rebasing the priority
// level to zero (the - MG_PRIORITY_HIGHEST, below), incrementing
// the rebased priority (+1), taking the modulus of the number of
// priorities to avoid exceeding the limit (% MG_NUM_PRIORITIES)
// and then restoring the base by adding back the first priority
// level (+ MG_PRIORITY_HIGHEST).
//
nextPriority = (((nextPriority + 1 - MG_PRIORITY_HIGHEST) %
MG_NUM_PRIORITIES) + MG_PRIORITY_HIGHEST);
}
//
// Use MGNewDataBuffer to allocate the buffer
//
rc = MGNewDataBuffer(pmgClient,
MG_TX_BUFFER,
bufferSize + sizeof(MG_INT_PKT_HEADER),
&pmgBufferArray[0]);
if (rc != 0)
{
WARNING_OUT(("MGNewDataBuffer failed in MGNewTxBuffer"));
DC_QUIT;
}
//
// Add the fields required for doing the send
//
pmgBufferArray[0]->priority = priority;
pmgBufferArray[0]->channelId = (ChannelID) channel;
pmgBufferArray[0]->senderId = pmgClient->userIDMCS;
ValidateFLOStr(pStr[0]);
pmgBufferArray[0]->pStr = pStr[0];
//
// Now allocate an additional three control blocks which are identical
// to the first one if required.
//
if (numPrioritiesToUse > 1)
{
//
// Firstly re-enable the NET_SEND_ALL_PRIORITIES flag. This is to
// ensure that traversing the linked list in MG_SendData is
// efficient.
//
pmgBufferArray[0]->priority |= NET_SEND_ALL_PRIORITIES;
//
// Create the duplicate buffers and initialise them.
//
for (i = 1; i < MG_NUM_PRIORITIES; i++)
{
TRACE_OUT(("Task allocating extra CB, priority %u",
priorities[i]));
//
// Allocate a new control buffer.
//
rc = MGNewBuffer(pmgClient,
MG_TX_BUFFER,
&pmgBufferArray[i]);
if (rc != 0)
{
WARNING_OUT(("MGNewBuffer failed"));
DC_QUIT;
}
//
// Initialise the buffer control block. The priority values of
// these control blocks are in increasing order from that of
// pmgBuffer.
//
pmgBufferArray[i]->priority = priorities[i];
pmgBufferArray[i]->channelId = pmgBufferArray[0]->channelId;
pmgBufferArray[i]->senderId = pmgBufferArray[0]->senderId;
pmgBufferArray[i]->length = pmgBufferArray[0]->length;
pmgBufferArray[i]->pPktHeader = pmgBufferArray[0]->pPktHeader;
pmgBufferArray[i]->pDataBuffer = pmgBufferArray[0]->pDataBuffer;
ValidateFLOStr(pStr[i]);
pmgBufferArray[i]->pStr = pStr[i];
//
// Set the NET_SEND_ALL_PRIORITIES flag.
//
pmgBufferArray[i]->priority |= NET_SEND_ALL_PRIORITIES;
//
// Now bump up the usage count of the data block.
//
pmgBufferArray[i]->pPktHeader->useCount++;
TRACE_OUT(("Use count of data buffer %#.8lx now %d",
pmgBufferArray[i]->pPktHeader,
pmgBufferArray[i]->pPktHeader->useCount));
}
}
//
// Assign the passed first control buffer allocated to the passed
// control buffer parameter.
//
*ppmgBuffer = pmgBufferArray[0];
DC_EXIT_POINT:
//
// In the event of a problem we free any buffers that we have already
// allocated.
//
if (rc != 0)
{
for (i = 0; i < MG_NUM_PRIORITIES; i++)
{
if (pmgBufferArray[i] != NULL)
{
TRACE_OUT(("About to free control buffer %u", i));
MGFreeBuffer(pmgClient, &pmgBufferArray[i]);
}
}
}
DebugExitDWORD(MGNewTxBuffer, rc);
return(rc);
}
//
// MGNewRxBuffer(...)
//
UINT MGNewRxBuffer
(
PMG_CLIENT pmgClient,
NET_PRIORITY priority,
NET_CHANNEL_ID channel,
NET_CHANNEL_ID senderID,
PMG_BUFFER * ppmgBuffer
)
{
UINT rc = 0;
DebugEntry(MGNewRxBuffer);
ValidateMGClient(pmgClient);
//
// First tell flow control we need a buffer.
// No back pressure may be applied here, but flow control uses this
// notification to control responses to the sender.
//
// Note that we always use the sizes including the internal packet
// header for flow control purposes.
//
FLO_AllocReceive(pmgClient,
priority,
channel,
senderID);
//
// Use MGNewDataBuffer to allocate the buffer. bufferSize includes the
// size of the network packet header (this comes over the wire), but
// not the remainder of the internal packet header.
//
rc = MGNewDataBuffer(pmgClient,
MG_RX_BUFFER,
sizeof(NET_SEND_IND_EVENT) + sizeof(MG_INT_PKT_HEADER),
ppmgBuffer);
//
// Add the fields required for a receive buffer
//
if (rc == 0)
{
(*ppmgBuffer)->priority = priority;
(*ppmgBuffer)->channelId = (ChannelID)channel;
(*ppmgBuffer)->senderId = (ChannelID)senderID;
}
else
{
WARNING_OUT(("MGNewDataBuffer failed in MGNewRxBuffer"));
}
DebugExitDWORD(MGNewRxBuffer, rc);
return(rc);
}
//
// MGFreeBuffer(...)
//
void MGFreeBuffer
(
PMG_CLIENT pmgClient,
PMG_BUFFER * ppmgBuffer
)
{
PMG_BUFFER pmgBuffer;
void * pBuffer;
DebugEntry(MGFreeBuffer);
pmgBuffer = *ppmgBuffer;
ValidateMGBuffer(pmgBuffer);
//
// If this is a receive buffer then we must first tell flow control
// about the space available
// This may trigger a pong, if we are waiting for the app to free up
// some space
//
if (pmgBuffer->type == MG_RX_BUFFER)
{
ASSERT (pmgBuffer->pPktHeader->useCount == 1);
TRACE_OUT(("Free RX pmgBuffer 0x%08x", pmgBuffer));
//
// Do a sanity check on the client (there is a window where this
// may have been freed).
//
if (!pmgClient->userAttached)
{
TRACE_OUT(("MGFreeBuffer: client 0x%08x not attached", pmgClient));
}
else
{
FLO_FreeReceive(pmgClient,
pmgBuffer->priority,
pmgBuffer->channelId,
pmgBuffer->senderId);
// Free the MCS buffer
if ((pmgBuffer->pPktHeader != NULL) && (pmgClient->m_piMCSSap != NULL))
{
ASSERT(pmgBuffer->pDataBuffer != NULL);
ASSERT(((PNET_SEND_IND_EVENT)pmgBuffer->pDataBuffer)->data_ptr != NULL);
pmgClient->m_piMCSSap->FreeBuffer (
(PVoid) (((PNET_SEND_IND_EVENT) pmgBuffer->pDataBuffer)
->data_ptr - sizeof(TSHR_NET_PKT_HEADER)));
TRACE_OUT(("MGFreeBuffer: Freed data_ptr for pmgBuffer 0x%08x pDataBuffer 0x%08x",
pmgBuffer, pmgBuffer->pDataBuffer));
((PNET_SEND_IND_EVENT)pmgBuffer->pDataBuffer)->data_ptr = NULL;
}
}
}
//
// Free the data buffer, if there is one present. Note that this can
// be referenced by more than one bufferCB, and so has a use count
// which must reach zero before the buffer is freed.
//
if (pmgBuffer->pPktHeader != NULL)
{
ASSERT(pmgBuffer->pPktHeader->useCount != 0);
pmgBuffer->pPktHeader->useCount--;
TRACE_OUT(("Data buffer 0x%08x use count %d",
pmgBuffer->pPktHeader,
pmgBuffer->pPktHeader->useCount));
if (pmgBuffer->pPktHeader->useCount == 0)
{
TRACE_OUT(("Freeing MG_BUFFER data 0x%08x; use count is zero", pmgBuffer->pPktHeader));
delete[] pmgBuffer->pPktHeader;
pmgBuffer->pPktHeader = NULL;
}
}
//
// If the buffer CB is in the pending queue then remove it first!
//
if (pmgBuffer->pendChain.next != 0)
{
COM_BasedListRemove(&(pmgBuffer->pendChain));
}
//
// Now remove the buffer CB itself from the list and free it up:
//
COM_BasedListRemove(&(pmgBuffer->clientChain));
delete pmgBuffer;
*ppmgBuffer = NULL;
DebugExitVOID(MGFreeBuffer);
}
//
// MGDetach(...)
//
void MGDetach
(
PMG_CLIENT pmgClient
)
{
PMG_BUFFER pmgBuffer;
PMG_BUFFER pmgT;
PIMCSSap pMCSSap;
#ifdef _DEBUG
UINT rc;
#endif // _DEBUG
DebugEntry(MGDetach);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
pMCSSap = pmgClient->m_piMCSSap;
//
// Remove any entries for this user from the channel join pending list.
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->joinChain),
FIELD_OFFSET(MG_BUFFER, pendChain));
while (pmgBuffer != NULL)
{
ValidateMGBuffer(pmgBuffer);
//
// Get a pointer to the next bufferCB in the list - we have to do
// this before we free the current bufferCB (freeing it NULLs it,
// so we won't be able to step along to the next entry in the
// list).
//
pmgT = (PMG_BUFFER)COM_BasedListNext(&(pmgClient->joinChain), pmgBuffer,
FIELD_OFFSET(MG_BUFFER, pendChain));
MGFreeBuffer(pmgClient, &pmgBuffer);
//
// We won't get a match on a join request now, so we don't have
// a join pending.
//
pmgClient->joinPending = FALSE;
pmgBuffer = pmgT;
}
//
// Remove any unsent receive buffers for this user from the buffer list
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->buffers),
FIELD_OFFSET(MG_BUFFER, clientChain));
while (pmgBuffer != NULL)
{
ValidateMGBuffer(pmgBuffer);
//
// Get a pointer to the next bufferCB in the list - we have to do
// this before we free the current bufferCB (freeing it NULLs it,
// so we won't be able to step along to the next entry in the
// list).
//
pmgT = (PMG_BUFFER)COM_BasedListNext(&(pmgClient->buffers), pmgBuffer,
FIELD_OFFSET(MG_BUFFER, clientChain));
if (pmgBuffer->type == MG_RX_BUFFER)
{
if (pmgBuffer->eventPosted)
{
if ((pmgBuffer->pPktHeader != NULL) && (pMCSSap != NULL))
{
ASSERT(pmgBuffer->pDataBuffer != NULL);
ASSERT(((PNET_SEND_IND_EVENT)pmgBuffer->pDataBuffer)->data_ptr != NULL);
pMCSSap->FreeBuffer (
(PVoid) (((PNET_SEND_IND_EVENT) pmgBuffer->pDataBuffer)
->data_ptr - sizeof(TSHR_NET_PKT_HEADER)));
TRACE_OUT(("MGDetach: Freed data_ptr for pmgBuffer 0x%08x pDataBuffer 0x%08x",
pmgBuffer, pmgBuffer->pDataBuffer));
((PNET_SEND_IND_EVENT) pmgBuffer->pDataBuffer)->data_ptr = NULL;
}
}
else
{
//
// The bufferCB's user matches the user we are freeing up,
// and we haven't posted the event to the user, so free it.
// MGFreeBuffer removes it from the pending list, so we don't
// have to do that.
//
MGFreeBuffer(pmgClient, &pmgBuffer);
}
}
pmgBuffer = pmgT;
}
//
// Clear out the attachment info
//
pmgClient->userAttached = FALSE;
pmgClient->userIDMCS = 0;
//
// We can safely do an MCS DetachRequest without adding a requestCB
// - MCS will not bounce the request due to congestion, domain merging
// etc.
//
if (pMCSSap != NULL)
{
#ifdef _DEBUG
rc = pMCSSap->ReleaseInterface();
if (rc != 0) {
//
// No quit - we need to do our own cleanup.
//
// lonchanc: what cleanup needs to be done???
//
rc = McsErrToNetErr(rc);
switch (rc)
{
case 0:
case NET_RC_MGC_INVALID_USER_HANDLE:
case NET_RC_MGC_TOO_MUCH_IN_USE:
// These are normal.
TRACE_OUT(("MCSDetachUser normal error %d", rc));
break;
default:
ERROR_OUT(("MCSDetachUser abnormal error %d", rc));
break;
}
}
#else
pMCSSap->ReleaseInterface();
#endif //_DEBUG
pmgClient->m_piMCSSap = NULL;
}
--g_mgAttachCount;
DebugExitVOID(MGDetach);
}
//
// MGProcessPendingQueue(...)
//
UINT MGProcessPendingQueue(PMG_CLIENT pmgClient)
{
PMG_BUFFER pmgBuffer;
PMG_BUFFER pNextBuffer;
UINT rc = 0;
DebugEntry(MGProcessPendingQueue);
ValidateMGClient(pmgClient);
pNextBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->pendChain),
FIELD_OFFSET(MG_BUFFER, pendChain));
//
// Try and clear all the pending request queue
//
for ( ; (pmgBuffer = pNextBuffer) != NULL; )
{
ValidateMGBuffer(pmgBuffer);
pNextBuffer = (PMG_BUFFER)COM_BasedListNext(&(pmgClient->pendChain),
pNextBuffer, FIELD_OFFSET(MG_BUFFER, pendChain));
TRACE_OUT(("Got request 0x%08x from queue, type %u",
pmgBuffer, pmgBuffer->type));
//
// Check that the buffer is still valid. There is a race at
// conference termination where we can arrive here, but our user
// has actually already detached. In this case, free the buffer
// and continue.
//
if (!pmgClient->userAttached)
{
TRACE_OUT(("MGProcessPendingQueue: client 0x%08x not attached", pmgClient));
MGFreeBuffer(pmgClient, &pmgBuffer);
continue;
}
switch (pmgBuffer->type)
{
case MG_RQ_CHANNEL_JOIN:
case MG_RQ_CHANNEL_JOIN_BY_KEY:
{
//
// If this client already has a join outstanding, then skip
// this request.
//
if (pmgClient->joinPending)
{
//
// Break out of switch and goto next iteration of for()
//
continue;
}
pmgClient->joinPending = TRUE;
//
// Attempt the join
//
rc = pmgClient->m_piMCSSap->ChannelJoin(
(unsigned short) pmgBuffer->channelId);
//
// If the join failed then post an error back immediately
//
if (rc != 0)
{
if ((rc != MCS_TRANSMIT_BUFFER_FULL) &&
(rc != MCS_DOMAIN_MERGING))
{
//
// Something terminal went wrong - post a
// NET_EV_JOIN_CONFIRM (failed) to the client
//
MGPostJoinConfirm(pmgClient,
NET_RESULT_USER_REJECTED,
pmgBuffer->channelId,
(NET_CHANNEL_ID)(pmgBuffer->work));
}
pmgClient->joinPending = FALSE;
}
else
{
//
// If the request worked then we must move it to the
// join queue for completion
//
TRACE_OUT(("Inserting 0x%08x into join queue",pmgBuffer));
COM_BasedListRemove(&(pmgBuffer->pendChain));
COM_BasedListInsertBefore(&(pmgClient->joinChain),
&(pmgBuffer->pendChain));
//
// Do not free this buffer - continue processing the
// pending queue
//
continue;
}
}
break;
case MG_RQ_CHANNEL_LEAVE:
{
//
// Try to leave the channel:
//
rc = pmgClient->m_piMCSSap->ChannelLeave(
(unsigned short) pmgBuffer->channelId);
if (rc == 0)
{
MGProcessEndFlow(pmgClient,
pmgBuffer->channelId);
}
}
break;
case MG_RQ_TOKEN_GRAB:
{
rc = pmgClient->m_piMCSSap->TokenGrab(pmgBuffer->channelId);
}
break;
case MG_RQ_TOKEN_INHIBIT:
{
rc = pmgClient->m_piMCSSap->TokenInhibit(pmgBuffer->channelId);
}
break;
case MG_RQ_TOKEN_RELEASE:
{
rc = pmgClient->m_piMCSSap->TokenRelease(pmgBuffer->channelId);
}
break;
case MG_TX_BUFFER:
{
ASSERT(!(pmgBuffer->pPktHeader->header.pktLength & TSHR_PKT_FLOW));
//
// Send the data. Remember that we don't send all of the
// packet header, only from the length...
//
ASSERT((pmgBuffer->priority != NET_TOP_PRIORITY));
rc = pmgClient->m_piMCSSap->SendData(NORMAL_SEND_DATA,
pmgBuffer->channelId,
(Priority)(pmgBuffer->priority),
(unsigned char *) &(pmgBuffer->pPktHeader->header),
pmgBuffer->pPktHeader->header.pktLength,
APP_ALLOCATION);
//
// Check the return code.
//
if (rc == 0)
{
//
// Update the allocation. FLO_DecrementAlloc will
// check that the stream pointer is not null for us.
// (It will be null if flow control has ended on this
// channel since this buffer was allocated or if this
// is an uncontrolled channel).
//
// Note that for flow control purposes, we always use
// packet sizes including the internal packet header.
//
FLO_DecrementAlloc(pmgBuffer->pStr,
(pmgBuffer->pPktHeader->header.pktLength
- sizeof(TSHR_NET_PKT_HEADER) + sizeof(MG_INT_PKT_HEADER)));
}
}
break;
case MG_TX_PING:
case MG_TX_PONG:
case MG_TX_PANG:
{
//
// This is the length of a ping/pong message:
//
ASSERT(pmgBuffer->priority != NET_TOP_PRIORITY);
rc = pmgClient->m_piMCSSap->SendData(NORMAL_SEND_DATA,
pmgBuffer->channelId,
(Priority)(pmgBuffer->priority),
(unsigned char *) &(pmgBuffer->pPktHeader->header),
sizeof(TSHR_NET_PKT_HEADER) + sizeof(TSHR_FLO_CONTROL),
APP_ALLOCATION);
}
break;
}
rc = McsErrToNetErr(rc);
//
// If the request failed due to back pressure then just get out
// now. We will try again later.
//
if (rc == NET_RC_MGC_TOO_MUCH_IN_USE)
{
TRACE_OUT(("MCS Back pressure"));
break;
}
//
// Only for obman...
//
if (pmgClient == &g_amgClients[MGTASK_OM])
{
ValidateCMP(g_pcmPrimary);
//
// For any other error or if everything worked so far
// then tell the user to keep going
//
TRACE_OUT(("Posting NET_FEEDBACK"));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_FEEDBACK,
0,
g_pcmPrimary->callID);
}
//
// All is OK, or the request failed fatally. In either case we
// should free this request and attempt to continue.
//
MGFreeBuffer(pmgClient, &pmgBuffer);
}
DebugExitDWORD(MGProcessPendingQueue, rc);
return(rc);
}
//
// MGPostJoinConfirm(...)
//
UINT MGPostJoinConfirm
(
PMG_CLIENT pmgClient,
NET_RESULT result,
NET_CHANNEL_ID channel,
NET_CHANNEL_ID correlator
)
{
PNET_JOIN_CNF_EVENT pNetJoinCnf;
PMG_BUFFER pmgBuffer;
UINT rc;
DebugEntry(MGPostJoinConfirm);
ValidateMGClient(pmgClient);
//
// Allocate a buffer to send the event in - this should only fail if we
// really are out of virtual memory.
//
rc = MGNewDataBuffer(pmgClient, MG_EV_BUFFER,
sizeof(MG_INT_PKT_HEADER) + sizeof(NET_JOIN_CNF_EVENT), &pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("Failed to alloc NET_JOIN_CNF_EVENT"));
DC_QUIT;
}
pNetJoinCnf = (PNET_JOIN_CNF_EVENT) pmgBuffer->pDataBuffer;
ValidateCMP(g_pcmPrimary);
if (!g_pcmPrimary->callID)
{
WARNING_OUT(("MGPostJoinConfirm failed; not in call"));
rc = NET_RC_MGC_NOT_CONNECTED;
DC_QUIT;
}
//
// Fill in the fields.
//
pNetJoinCnf->callID = g_pcmPrimary->callID;
pNetJoinCnf->result = result;
pNetJoinCnf->channel = channel;
pNetJoinCnf->correlator = correlator;
//
// OK, we've built the event so now post it to our client:
//
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_EVENT_CHANNEL_JOIN,
0,
(UINT_PTR) pNetJoinCnf);
pmgBuffer->eventPosted = TRUE;
DC_EXIT_POINT:
DebugExitDWORD(MGPostJoinConfirm, rc);
return(rc);
}
//
// MCSErrToNetErr()
//
UINT McsErrToNetErr ( UINT rcMCS )
{
UINT rc = NET_RC_MGC_NOT_SUPPORTED;
//
// We use a static array of values to map the return code:
//
if (rcMCS < sizeof(c_RetCodeMap1) / sizeof(c_RetCodeMap1[0]))
{
rc = c_RetCodeMap1[rcMCS];
}
else
{
UINT nNewIndex = rcMCS - MCS_DOMAIN_ALREADY_EXISTS;
if (nNewIndex < sizeof(c_RetCodeMap2) / sizeof(c_RetCodeMap2[0]))
{
rc = c_RetCodeMap2[nNewIndex];
}
}
#ifdef _DEBUG
if (MCS_TRANSMIT_BUFFER_FULL == rcMCS)
{
ASSERT(NET_RC_MGC_TOO_MUCH_IN_USE == rc);
}
#endif
return rc;
}
//
// TranslateResult(...)
//
NET_RESULT TranslateResult(WORD resultMCS)
{
//
// We use a static array of values to map the result code:
//
if (resultMCS >= MG_NUM_OF_MCS_RESULTS)
resultMCS = MG_INVALID_MCS_RESULT;
return(c_ResultMap[resultMCS]);
}
//
// MGFLOCallback(...)
//
void MGFLOCallBack
(
PMG_CLIENT pmgClient,
UINT callbackType,
UINT priority,
UINT newBufferSize
)
{
PMG_BUFFER pmgBuffer;
DebugEntry(MGFLOCallBack);
ASSERT(priority != NET_TOP_PRIORITY);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
//
// If this is a buffermod callback then tell the app
//
if (pmgClient == &g_amgClients[MGTASK_DCS])
{
if (callbackType == FLO_BUFFERMOD)
{
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_FLOW,
priority,
newBufferSize);
}
}
else
{
ASSERT(pmgClient == &g_amgClients[MGTASK_OM]);
//
// Wake up the app in case we have applied back pressure.
//
TRACE_OUT(("Posting NET_FEEDBACK"));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_FEEDBACK,
0,
g_pcmPrimary->callID);
}
DebugExitVOID(MGFLOCallback);
}
//
// MGProcessDomainWatchdog()
//
void MGProcessDomainWatchdog
(
PMG_CLIENT pmgClient
)
{
int task;
DebugEntry(MGProcessDomainWatchdog);
ValidateMGClient(pmgClient);
//
// Call FLO to check each user attachment for delinquency
//
if (g_mgAttachCount > 0)
{
for (task = MGTASK_FIRST; task < MGTASK_MAX; task++)
{
if (g_amgClients[task].userAttached)
{
FLO_CheckUsers(&(g_amgClients[task]));
}
}
//
// Continue periodic messages - but only if there are some users.
//
// TRACE_OUT(("Continue watchdog"));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
MG_TIMER_PERIOD,
NET_MG_WATCHDOG,
0, 0);
}
else
{
TRACE_OUT(("Don't continue Watchdog timer"));
}
DebugExitVOID(MGProcessDomainWatchdog);
}
//
// FLO_UserTerm
//
void FLO_UserTerm(PMG_CLIENT pmgClient)
{
UINT i;
UINT cStreams;
DebugEntry(FLO_UserTerm);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
cStreams = pmgClient->flo.numStreams;
//
// Stop flow control on all channels. We scan the list of streams and
// if flow control is active on a stream then we stop it.
//
for (i = 0; i < cStreams; i++)
{
//
// Check that the stream is flow controlled.
//
if (pmgClient->flo.pStrData[i] != NULL)
{
//
// End control on this controlled stream.
//
FLOStreamEndControl(pmgClient, i);
}
}
DebugExitVOID(FLO_UserTerm);
}
//
// FLO_StartControl
//
void FLO_StartControl
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID channel,
UINT priority,
UINT backlog,
UINT maxBytesOutstanding
)
{
UINT rc = 0;
PFLO_STREAM_DATA pStr;
UINT i;
UINT stream;
DebugEntry(FLO_StartControl);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Flow control is on by default.
//
//
// Check to see if the channel is already flow controlled. If it is
// then we just exit.
//
stream = FLOGetStream(pmgClient, channel, priority, &pStr);
if (stream != FLO_NOT_CONTROLLED)
{
ValidateFLOStr(pStr);
TRACE_OUT(("Stream %u is already controlled (0x%08x:%u)",
stream, channel, priority));
DC_QUIT;
}
//
// If we already have hit the stream limit for this app then give up.
//
for (i = 0; i < FLO_MAX_STREAMS; i++)
{
if ((pmgClient->flo.pStrData[i]) == NULL)
{
break;
}
}
if (i == FLO_MAX_STREAMS)
{
ERROR_OUT(("Too many streams defined already"));
DC_QUIT;
}
TRACE_OUT(("This is stream %u", i));
//
// Allocate memory for our stream data. Hang the pointer off floHandle
// - this should be returned to us on all subsequent API calls.
//
pStr = new FLO_STREAM_DATA;
if (!pStr)
{
WARNING_OUT(("FLO_StartControl failed; out of memory"));
DC_QUIT;
}
ZeroMemory(pStr, sizeof(*pStr));
//
// Store the channel and priorities for this stream.
//
SET_STAMP(pStr, FLOSTR);
pStr->channel = channel;
pStr->priority = priority;
pStr->backlog = backlog;
if (maxBytesOutstanding == 0)
{
maxBytesOutstanding = FLO_MAX_STREAMSIZE;
}
pStr->DC_ABSMaxBytesInPipe = maxBytesOutstanding;
pStr->maxBytesInPipe = FLO_INIT_STREAMSIZE;
if (pStr->maxBytesInPipe > maxBytesOutstanding)
{
pStr->maxBytesInPipe = maxBytesOutstanding;
}
//
// Set the initial stream bytesAllocated to 0.
//
pStr->bytesAllocated = 0;
//
// Ping needed immediately.
//
pStr->pingNeeded = TRUE;
pStr->pingTime = FLO_INIT_PINGTIME;
pStr->nextPingTime = GetTickCount();
//
// Initialize the users base pointers.
//
COM_BasedListInit(&(pStr->users));
//
// Hang the stream CB off the base control block.
//
pmgClient->flo.pStrData[i] = pStr;
if (i >= pmgClient->flo.numStreams)
{
pmgClient->flo.numStreams++;
}
TRACE_OUT(("Flow control started, stream %u, (0x%08x:%u)",
i, channel, priority));
DC_EXIT_POINT:
DebugExitVOID(FLO_StartControl);
}
//
// FLO_EndControl
//
void FLO_EndControl
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID channel,
UINT priority
)
{
UINT stream;
PFLO_STREAM_DATA pStr;
DebugEntry(FLO_EndControl);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Convert channel and stream into priority.
//
stream = FLOGetStream(pmgClient, channel, priority, &pStr);
//
// The stream is not controlled so we just trace and quit.
//
if (stream == FLO_NOT_CONTROLLED)
{
WARNING_OUT(("Uncontrolled stream channel 0x%08x priority %u",
channel, priority));
DC_QUIT;
}
//
// Call the internal FLOStreamEndControl to end flow control on a
// given stream.
//
ValidateFLOStr(pStr);
FLOStreamEndControl(pmgClient, stream);
DC_EXIT_POINT:
DebugExitVOID(FLO_EndControl);
}
//
// FLO_AllocSend
//
UINT FLO_AllocSend
(
PMG_CLIENT pmgClient,
UINT priority,
NET_CHANNEL_ID channel,
UINT size,
PFLO_STREAM_DATA * ppStr
)
{
UINT stream;
UINT curtime;
PFLO_STREAM_DATA pStr;
BOOL denyAlloc = FALSE;
BOOL doPing = FALSE;
UINT rc = 0;
DebugEntry(FLO_AllocSend);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Convert channel and stream into priority
//
stream = FLOGetStream(pmgClient, channel, priority, ppStr);
pStr = *ppStr;
//
// For non-controlled streams just send the data
//
if (stream == FLO_NOT_CONTROLLED)
{
TRACE_OUT(("Send %u bytes on uncontrolled channel/pri (0x%08x:%u)",
size, channel, priority));
DC_QUIT;
}
//
// Get the current tick count.
//
curtime = GetTickCount();
//
// Check whether this request is permitted. We must allow one packet
// beyond the specified limit to avoid problems determining when we
// have started rejecting requests and also to avoid situations where a
// single request exceeds the total pipe size.
//
// If we have not yet received a pong then we limit the amount of
// allocated buffer space to below FLO_MAX_PRE_FC_ALLOC. However this
// data can be sent immediately so the overall throughput is still
// relatively high. In this way we minimize the amount of data held in
// the glue layer to a maximum of FLO_MAX_PRE_FC_ALLOC if there are no
// remote users.
//
ValidateFLOStr(pStr);
if (!pStr->gotPong)
{
//
// Flag that a ping is required.
//
pStr->pingNeeded = TRUE;
if (curtime > pStr->nextPingTime)
{
doPing = TRUE;
}
//
// We haven't got a pong yet (i.e. FC is non-operational) so we
// need to limit the maximum amount of data held in flow control to
// FLO_MAX_PRE_FC_ALLOC.
//
if (pStr->bytesAllocated > FLO_MAX_PRE_FC_ALLOC)
{
denyAlloc = TRUE;
TRACE_OUT(("Max allocation of %u bytes exceeded (currently %u)",
FLO_MAX_PRE_FC_ALLOC,
pStr->bytesAllocated));
DC_QUIT;
}
pStr->bytesInPipe += size;
pStr->bytesAllocated += size;
TRACE_OUT((
"Alloc of %u succeeded: bytesAlloc %u, bytesInPipe %u"
" (0x%08x:%u)",
size,
pStr->bytesAllocated,
pStr->bytesInPipe,
pStr->channel,
pStr->priority));
DC_QUIT;
}
if (pStr->bytesInPipe < pStr->maxBytesInPipe)
{
//
// Check to see if a ping is required and if so send it now.
//
if ((pStr->pingNeeded) && (curtime > pStr->nextPingTime))
{
doPing = TRUE;
}
pStr->bytesInPipe += size;
pStr->bytesAllocated += size;
TRACE_OUT(("Stream %u - alloc %u (InPipe:MaxInPipe %u:%u)",
stream,
size,
pStr->bytesInPipe,
pStr->maxBytesInPipe));
DC_QUIT;
}
//
// If we get here then we cannot currently allocate any buffers so deny
// the allocation. Simulate back pressure with NET_OUT_OF_RESOURCE.
// We also flag that a "wake up" event is required to get the app to
// send more data.
//
denyAlloc = TRUE;
pStr->eventNeeded = TRUE;
pStr->curDenialTime = pStr->lastPingTime;
//
// We are not allowed to apply back pressure unless we can guarantee
// that we will wake up the app later on. This is dependent upon our
// receiving a pong later. But if there is no ping outstanding
// (because we have allocated all our buffer allowance within the ping
// delay time) then we should first send a ping to trigger the wake up.
// If this fails then our watchdog will finally wake us up.
//
if (pStr->pingNeeded)
{
doPing = TRUE;
}
DC_EXIT_POINT:
//
// Check to see if we should deny the buffer allocation.
//
if (denyAlloc)
{
rc = NET_RC_MGC_TOO_MUCH_IN_USE;
TRACE_OUT(("Denying buffer request on stream %u InPipe %u Alloc %u",
stream,
pStr->bytesInPipe,
pStr->bytesAllocated));
}
if (doPing)
{
//
// A ping is required so send it now.
//
FLOPing(pmgClient, stream, curtime);
}
DebugExitDWORD(FLO_AllocSend, rc);
return(rc);
}
//
// FLO_ReallocSend
//
void FLO_ReallocSend
(
PMG_CLIENT pmgClient,
PFLO_STREAM_DATA pStr,
UINT size
)
{
DebugEntry(FLO_ReallocSend);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
//
// For non-controlled streams there is nothing to do so just exit.
//
if (pStr == NULL)
{
TRACE_OUT(("Realloc data on uncontrolled channel"));
DC_QUIT;
}
//
// Perform a quick sanity check.
//
ValidateFLOStr(pStr);
if (size > pStr->bytesInPipe)
{
ERROR_OUT(("Realloc of %u makes bytesInPipe (%u) neg (0x%08x:%u)",
size,
pStr->bytesInPipe,
pStr->channel,
pStr->priority));
DC_QUIT;
}
//
// Add the length not sent back into the pool.
//
pStr->bytesInPipe -= size;
TRACE_OUT(("Realloc %u FC bytes (bytesInPipe is now %u) (0x%08x:%u)",
size,
pStr->bytesInPipe,
pStr->channel,
pStr->priority));
DC_EXIT_POINT:
//
// Every time that we call FLO_ReallocSend we also want to call
// FLO_DecrementAlloc (but not vice-versa) so call it now.
//
FLO_DecrementAlloc(pStr, size);
DebugExitVOID(FLO_ReallocSend);
}
//
// FLO_DecrementAlloc
//
void FLO_DecrementAlloc
(
PFLO_STREAM_DATA pStr,
UINT size
)
{
DebugEntry(FLO_DecrementAlloc);
//
// For non-controlled streams there is nothing to do so just exit.
//
if (pStr == NULL)
{
TRACE_OUT(("Decrement bytesAllocated on uncontrolled channel"));
DC_QUIT;
}
//
// Perform a quick sanity check.
//
ValidateFLOStr(pStr);
if (size > pStr->bytesAllocated)
{
ERROR_OUT(("Dec of %u makes bytesAllocated (%u) neg (0x%08x:%u)",
size,
pStr->bytesAllocated,
pStr->channel,
pStr->priority));
DC_QUIT;
}
//
// Update the count of the data held in the glue for this stream.
//
pStr->bytesAllocated -= size;
TRACE_OUT(("Clearing %u alloc bytes (bytesAlloc is now %u) (0x%08x:%u)",
size,
pStr->bytesAllocated,
pStr->channel,
pStr->priority));
DC_EXIT_POINT:
DebugExitVOID(FLO_DecrementAlloc);
}
//
// FLO_CheckUsers
//
void FLO_CheckUsers(PMG_CLIENT pmgClient)
{
PFLO_USER pFloUser;
PBASEDLIST nextUser;
int waited;
BYTE stream;
UINT curtime;
PFLO_STREAM_DATA pStr;
DebugEntry(FLO_CheckUsers);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
curtime = GetTickCount();
//
// Check users of each stream
//
for (stream = 0; stream < pmgClient->flo.numStreams; stream++)
{
if (pmgClient->flo.pStrData[stream] == NULL)
{
continue;
}
pStr = pmgClient->flo.pStrData[stream];
ValidateFLOStr(pStr);
//
// Check whether we have waited long enough and need to reset the
// wait counters. We only wait a certain time before resetting all
// our counts. What has happened is that someone has left the call
// and we have been waiting for their pong.
//
// We detect the outage by checking against nextPingTime which, as
// well as being set to the earliest time we can send a ping is
// also updated to the current time as each pong comes in so we can
// use it as a measure of the time since the last repsonse from any
// user of the stream.
//
// To avoid false outages caused by new joiners or transient large
// buffer situations each user is required to send a pong at the
// rate of MAX_WAIT_TIME/2. They do this by just sending a
// duplicate pong if they have not yet got the ping they need to
// to pong.
//
if ((pStr->eventNeeded) &&
(!pStr->pingNeeded))
{
TRACE_OUT(("Checking for valid back pressure on stream %u",
stream));
//
// Note that if there are no remote users then we should reset
// the flags regardless. We get into this state when we first
// start an app because OBMAN sends data before the app has
// joined the channel at the other end.
//
waited = curtime - pStr->nextPingTime;
if (waited > FLO_MAX_WAIT_TIME)
{
TRACE_OUT(("Stream %u - Waited for %d, resetting counter",
stream, waited));
pStr->bytesInPipe = 0;
pStr->pingNeeded = TRUE;
pStr->nextPingTime = curtime;
pStr->gotPong = FALSE;
//
// Remove outdated records from our user queue
//
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pStr->users));
while (&(pFloUser->list) != &(pStr->users))
{
ValidateFLOUser(pFloUser);
//
// Address the follow on record before we free the
// current
//
nextUser = COM_BasedNextListField(&(pFloUser->list));
//
// Free the current record, if necessary
//
if (pFloUser->lastPongRcvd != pStr->pingValue)
{
//
// Remove from the list
//
TRACE_OUT(("Freeing FLO_USER 0x%08x ID 0x%08x", pFloUser, pFloUser->userID));
COM_BasedListRemove(&(pFloUser->list));
delete pFloUser;
}
else
{
//
// At least one user still out there so keep flow
// control active or else we would suddenly send
// out a burst of data that might flood them
//
pStr->gotPong = TRUE;
}
//
// Move on to the next record in the list
//
pFloUser = (PFLO_USER)nextUser;
}
//
// We have previously rejected an application request so we
// had better call back now
//
if (pmgClient->flo.callBack != NULL)
{
(*(pmgClient->flo.callBack))(pmgClient,
FLO_WAKEUP,
pStr->priority,
pStr->maxBytesInPipe);
}
pStr->eventNeeded = FALSE;
}
}
}
DebugExitVOID(FLO_CheckUsers);
}
//
// FLO_ReceivedPacket
//
void FLO_ReceivedPacket
(
PMG_CLIENT pmgClient,
PTSHR_FLO_CONTROL pPkt
)
{
BOOL canPing = TRUE;
PFLO_USER pFloUser;
BOOL userFound = FALSE;
UINT stream;
UINT curtime;
PFLO_STREAM_DATA pStr;
UINT callbackType = 0;
int latency;
UINT throughput;
DebugEntry(FLO_ReceivedPacket);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
stream = pPkt->stream;
ASSERT(stream < FLO_MAX_STREAMS);
pStr = pmgClient->flo.pStrData[stream];
//
// If the stream CB has been freed up already then we can ignore any
// flow information pertaining to it.
//
if (pStr == NULL)
{
TRACE_OUT(("Found a null stream pointer for stream %u", stream));
DC_QUIT;
}
ValidateFLOStr(pStr);
curtime = GetTickCount();
//
// First we must locate the user for this ping/pong/pang
// Also, while we are doing it we can check to see if it is a pong and
// if so whether it is the last pong we need
//
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pStr->users));
while (&(pFloUser->list) != &(pStr->users))
{
ValidateFLOUser(pFloUser);
if (pFloUser->userID == pPkt->userID)
{
userFound = TRUE;
//
// We have got a match so set up the last pong value
// Accumulate pong stats for query
//
if (pPkt->packetType == PACKET_TYPE_PONG)
{
pFloUser->lastPongRcvd = pPkt->pingPongID;
pFloUser->gotPong = TRUE;
pFloUser->numPongs++;
pFloUser->pongDelay += curtime - pStr->lastPingTime;
}
else
{
break;
}
}
//
// So, is it the final pong - are there any users with different
// pong required entries?
// Note that if the user has never sent us a pong then we don't
// reference their lastPongRcvd field at this stage.
//
if (pPkt->packetType == PACKET_TYPE_PONG)
{
if (pFloUser->gotPong &&
(pFloUser->lastPongRcvd != pStr->pingValue))
{
TRACE_OUT(("%u,%u - Entry 0x%08x has different ping id %u",
stream,
pFloUser->userID,
pFloUser,
pFloUser->lastPongRcvd));
canPing = FALSE;
}
}
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pFloUser->list));
}
//
// If this is a new User then add them to the list
//
if (!userFound)
{
pFloUser = FLOAddUser(pPkt->userID, pStr);
//
// If this is a pong then we can set up the lastpong as well
//
if ((pFloUser != NULL) &&
(pPkt->packetType == PACKET_TYPE_PONG))
{
pFloUser->lastPongRcvd = pPkt->pingPongID;
}
}
//
// Now perform the actual packet specific processing
//
switch (pPkt->packetType)
{
//
// PING
//
// If this is a ping packet then just flag we must send a pong. If
// we failed to alloc a user CB then just ignore the ping and they
// will continue in blissful ignorance of our presence
//
case PACKET_TYPE_PING:
{
TRACE_OUT(("%u,%u - PING %u received",
stream, pPkt->userID, pPkt->pingPongID));
ValidateFLOUser(pFloUser);
pFloUser->sendPongID = pPkt->pingPongID;
if (pFloUser->rxPackets < FLO_MAX_RCV_PACKETS)
{
FLOPong(pmgClient, stream, pFloUser->userID, pPkt->pingPongID);
pFloUser->sentPongTime = curtime;
}
else
{
TRACE_OUT(("Receive backlog - just flagging pong needed"));
pFloUser->pongNeeded = TRUE;
}
}
break;
//
// PONG
//
// Flag we have got a pong from any user so we should start
// applying send flow control to this stream now (Within the stream
// we achieve per user granularity by ignoring those users that
// have never ponged when we inspect the stream byte count.)
//
case PACKET_TYPE_PONG:
{
pStr->gotPong = TRUE;
//
// Keep a note that we are receiving messages on this stream by
// moving nextPing on (but only if we have passed it)
//
if (curtime > pStr->nextPingTime)
{
pStr->nextPingTime = curtime;
}
//
// Update the user entry and schedule a ping if necessary
//
TRACE_OUT(("%u,%u - PONG %u received",
stream, pPkt->userID, pPkt->pingPongID));
//
// Check for readiness to send another ping This may be because
// this is the first users pong, in which case we should also send
// another ping when ready
//
if (canPing)
{
TRACE_OUT(("%u - PING scheduled, pipe was %d",
stream,
pStr->bytesInPipe));
//
// Reset byte count and ping readiness flag
//
pStr->bytesInPipe = 0;
pStr->pingNeeded = TRUE;
//
// Adjust the buffer size limit based on our current throughput
//
// If we hit the back pressure point and yet we are ahead of
// the target backlog then we should increase the buffer size
// to avoid constraining the pipe. If we have already
// increased the buffer size to our maximum value then try
// decreasing the tick delay. If we are already ticking at the
// max rate then we are going as fast as we can. If we make
// either of these adjustments then allow the next ping to flow
// immediately so that we can ramp up as fast as possible to
// LAN bandwidths.
//
// We dont need to do the decrease buffer checks if we have not
// gone into back pressure during the last pong cycle
//
if (pStr->eventNeeded)
{
TRACE_OUT(("We were in a back pressure situation"));
callbackType = FLO_WAKEUP;
TRACE_OUT(("Backlog %u denial delta %d ping delta %d",
pStr->backlog, curtime-pStr->lastDenialTime,
curtime-pStr->lastPingTime));
//
// The next is a little complex.
//
// If the turnaround of this ping pong is significantly
// less than our target then open the pipe up. But we must
// adjust to allow for the ping being sent at a quiet
// period, which we do by remembering when each ping is
// sent and, if we encounter a backlog situation, storing
// that ping time for future reference
//
// So the equation for latency is
//
// Pongtime-previous backlogged ping time
//
// The previous ping time is the that we sent prior to the
// last back pressure situation so there are two times in
// the control block, one for the last Ping time and one
// for the last but one ping time.
//
if ((int)(pStr->backlog/2 - curtime +
pStr->lastDenialTime) > 0)
{
//
// We are coping easily so increase the buffer to pump
// more data through. Predict the new buffer size
// based on the latency for the current backlog so that
// we don't artificially constrain the app. We do this
// by taking the observed latency, decrementing by a
// small factor to allow for the latency we might
// observe over the fastest possible link and then
// calculating the connection throughput.
//
// latency = curtime - lastDenialTime - fudge(100mS)
// amount sent = maxBytesInPipe (because we we were
// backed up)
// throughput = amount sent/latency (bytes/millisec)
// New buffer = throughput * target latency
//
if (pStr->maxBytesInPipe < pStr->DC_ABSMaxBytesInPipe)
{
latency = (curtime -
pStr->lastDenialTime -
30);
if (latency <= 0)
{
latency = 1;
}
throughput = (pStr->maxBytesInPipe*8)/latency;
pStr->maxBytesInPipe = (throughput * pStr->backlog)/8;
TRACE_OUT(("Potential maxbytes of %d",
pStr->maxBytesInPipe));
if (pStr->maxBytesInPipe > pStr->DC_ABSMaxBytesInPipe)
{
pStr->maxBytesInPipe = pStr->DC_ABSMaxBytesInPipe;
}
TRACE_OUT((
"Modified buffer maxBytesInPipe up to %u "
"(0x%08x:%u)",
pStr->maxBytesInPipe,
pStr->channel,
pStr->priority));
callbackType = FLO_BUFFERMOD;
}
else
{
//
// We have hit our maximum allowed pipe size but
// are still backlogged and yet pings are going
// through acceptably.
//
// Our first action is to try reducing the ping
// time thus increasing out throughput.
//
// If we have already decreased the ping time to
// its minimum then we cannot do anything else. It
// is possible that the application parameters
// should be changed to increase the permissible
// throughput so log an alert to suggest this.
// however there are situations (input management)
// where we want some back pressure in order to
// prevent excessive cpu loading at the recipient.
//
// To increase the throughput either
//
// - Increase the maximum size of the stream. The
// disadvantage of this is that a low badwidth
// joiner may suddenly see a lot of high
// bandwidth data in the pipe. However this
// is the preferred solution in general, as
// it avoids having the pipe flooded with pings.
//
// - Reduce the target latency. This is a little
// dangerous because the latency is composed of
// the pre-queued data and the network turnaround
// time and if the network turnaround time
// approaches the target latency then the flow
// control will simply close the pipe right down
// irrespective of the achievable throughput.
//
pStr->maxBytesInPipe = pStr->DC_ABSMaxBytesInPipe;
pStr->pingTime = pStr->pingTime/2;
if (pStr->pingTime < FLO_MIN_PINGTIME)
{
pStr->pingTime = FLO_MIN_PINGTIME;
}
TRACE_OUT((
"Hit DC_ABS max - reduce ping time to %u",
pStr->pingTime));
}
//
// Allow the ping just scheduled to flow immediately
//
pStr->nextPingTime = curtime;
}
pStr->eventNeeded = FALSE;
}
//
// If we have exceeded our target latency at all then throttle
// back
//
if ((int)(pStr->backlog - curtime + pStr->lastPingTime) < 0)
{
pStr->maxBytesInPipe /= 2;
if (pStr->maxBytesInPipe < FLO_MIN_STREAMSIZE)
{
pStr->maxBytesInPipe = FLO_MIN_STREAMSIZE;
}
pStr->pingTime = pStr->pingTime * 2;
if (pStr->pingTime > FLO_INIT_PINGTIME)
{
pStr->pingTime = FLO_INIT_PINGTIME;
}
TRACE_OUT((
"Mod buffer maxBytesInPipe down to %u, ping to %u "
"(0x%08x:%u)",
pStr->maxBytesInPipe,
pStr->pingTime,
pStr->channel,
pStr->priority));
callbackType = FLO_BUFFERMOD;
}
//
// Now make athe callback if callbackType has been set
//
if ((callbackType != 0) &&
(pmgClient->flo.callBack != NULL))
{
(pmgClient->flo.callBack)(pmgClient,
callbackType,
pStr->priority,
pStr->maxBytesInPipe);
}
}
}
break;
//
// PANG
//
// Remove the user and continue
//
case PACKET_TYPE_PANG:
{
TRACE_OUT(("%u,%u - PANG received, removing user",
stream, pPkt->userID));
//
// Remove from the list
//
ValidateFLOUser(pFloUser);
TRACE_OUT(("Freeing FLO_USER 0x%08x ID 0x%08x", pFloUser, pFloUser->userID));
COM_BasedListRemove(&(pFloUser->list));
delete pFloUser;
//
// If we are currently waiting then generate an event for the
// app to get it moving again
//
if ((pStr->eventNeeded) &&
(pmgClient->flo.callBack != NULL))
{
TRACE_OUT(("Waking up the app because user has left"));
(*(pmgClient->flo.callBack))(pmgClient,
FLO_WAKEUP,
pStr->priority,
pStr->maxBytesInPipe);
pStr->eventNeeded = FALSE;
}
}
break;
//
// UNKNOWN
//
// Just trace alert and press on
//
default:
{
WARNING_OUT(("Invalid packet type 0x%08x", pPkt->packetType));
}
break;
}
DC_EXIT_POINT:
DebugExitVOID(FLO_ReceivedPacket);
}
//
// FLO_AllocReceive
//
void FLO_AllocReceive
(
PMG_CLIENT pmgClient,
UINT priority,
NET_CHANNEL_ID channel,
UINT userID
)
{
UINT stream;
PFLO_USER pFloUser;
BOOL userFound = FALSE;
PFLO_STREAM_DATA pStr;
UINT curtime;
DebugEntry(FLO_AllocReceive);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Convert channel and priority into stream
//
stream = FLOGetStream(pmgClient, channel, priority, &pStr);
//
// Only process controlled streams
//
if (stream == FLO_NOT_CONTROLLED)
{
DC_QUIT;
}
//
// First we must locate the user
//
ValidateFLOStr(pStr);
COM_BasedListFind(LIST_FIND_FROM_FIRST, &(pStr->users),
(void**)&pFloUser, FIELD_OFFSET(FLO_USER, list), FIELD_OFFSET(FLO_USER, userID),
(DWORD)userID, FIELD_SIZE(FLO_USER, userID));
//
// SFR6101: If this is a new User then add them to the list
//
if (pFloUser == NULL)
{
TRACE_OUT(("Message from user 0x%08x who is not flow controlled", userID));
pFloUser = FLOAddUser(userID, pStr);
}
//
// If we failed to allocate a usr CB then just ignore for now
//
if (pFloUser != NULL)
{
ValidateFLOUser(pFloUser);
//
// Add in the new receive packet usage
//
pFloUser->rxPackets++;
TRACE_OUT(("Num outstanding receives on stream %u now %u",
stream, pFloUser->rxPackets));
//
// Now check that we have not got some kind of creep
//
if (pFloUser->rxPackets > FLO_MAX_RCV_PKTS_CREEP)
{
WARNING_OUT(("Creep? Stream %u has %u unacked rcv pkts",
stream, pFloUser->rxPackets));
}
//
// Finally check to see that we are responding OK to this person
//
curtime = GetTickCount();
if ((pFloUser->pongNeeded) &&
(curtime - pFloUser->sentPongTime > (FLO_MAX_WAIT_TIME/4)))
{
TRACE_OUT(("Send keepalive pong"));
FLOPong(pmgClient, stream, pFloUser->userID, pFloUser->sendPongID);
pFloUser->sentPongTime = curtime;
}
}
DC_EXIT_POINT:
DebugExitVOID(FLO_AllocReceive);
}
//
// FLO_FreeReceive
//
void FLO_FreeReceive
(
PMG_CLIENT pmgClient,
NET_PRIORITY priority,
NET_CHANNEL_ID channel,
UINT userID
)
{
UINT stream;
PFLO_USER pFloUser;
PFLO_STREAM_DATA pStr;
BOOL userFound = FALSE;
DebugEntry(FLO_FreeReceive);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
//
// Convert channel and priority into stream
//
stream = FLOGetStream(pmgClient, channel, priority, &pStr);
//
// Only process controlled streams
//
if (stream != FLO_NOT_CONTROLLED)
{
ValidateFLOStr(pStr);
//
// First we must locate the user
//
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pStr->users));
while (&(pFloUser->list) != &(pStr->users))
{
ValidateFLOUser(pFloUser);
if (pFloUser->userID == userID)
{
userFound = TRUE;
break;
}
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pFloUser->list));
}
//
// If we do not find the user record then two things may have
// happened.
// - They have joined the channel and immediately sent data
// - They were removed as being delinquent and are now sending
// data again
// - We failed to add them to our user list
// Try and allocate the user entry now
// (This will start tracking receive buffer space, but this user
// will not participate in our send flow control until we receive
// a pong from them and set "gotpong" in their FLO_USER CB.)
//
if (!userFound)
{
pFloUser = FLOAddUser(userID, pStr);
}
if (pFloUser != NULL)
{
ValidateFLOUser(pFloUser);
//
// Check that we have not got some kind of creep
//
if (pFloUser->rxPackets == 0)
{
WARNING_OUT(("Freed too many buffers for user 0x%08x on str %u",
userID, stream));
}
else
{
pFloUser->rxPackets--;
TRACE_OUT(("Num outstanding receives now %u",
pFloUser->rxPackets));
}
//
// Now we must Pong if there is a pong pending and we have
// moved below the high water mark
//
if ((pFloUser->pongNeeded) &&
(pFloUser->rxPackets < FLO_MAX_RCV_PACKETS))
{
FLOPong(pmgClient, stream, pFloUser->userID, pFloUser->sendPongID);
pFloUser->pongNeeded = FALSE;
pFloUser->sentPongTime = GetTickCount();
}
}
}
DebugExitVOID(FLO_FreeReceive);
}
//
// FLOPong()
//
void FLOPong
(
PMG_CLIENT pmgClient,
UINT stream,
UINT userID,
UINT pongID
)
{
PTSHR_FLO_CONTROL pFlo;
PMG_BUFFER pmgBuffer;
UINT rc;
DebugEntry(FLOPong);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
rc = MGNewDataBuffer(pmgClient,
MG_TX_PONG,
sizeof(TSHR_FLO_CONTROL) + sizeof(MG_INT_PKT_HEADER),
&pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewDataBuffer failed in FLOPong"));
DC_QUIT;
}
pFlo = (PTSHR_FLO_CONTROL)pmgBuffer->pDataBuffer;
pmgBuffer->pPktHeader->header.pktLength = TSHR_PKT_FLOW;
//
// Set up pong contents
//
pFlo->packetType = PACKET_TYPE_PONG;
pFlo->userID = pmgClient->userIDMCS;
pFlo->stream = (BYTE)stream;
pFlo->pingPongID = (BYTE)pongID;
pmgBuffer->channelId = (ChannelID)userID;
pmgBuffer->priority = MG_PRIORITY_HIGHEST;
//
// Now decouple the send request. Note that we must put the pong at
// the back of the request queue even though we want it to flow at
// high priority because otherwise there are certain circumstances
// where we get pong reversal due to receipt of multiple pings
//
TRACE_OUT(("Inserting pong message 0x%08x at head of pending chain", pmgBuffer));
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
TRACE_OUT(("%u,0x%08x - PONG %u scheduled",
pFlo->stream, pmgBuffer->channelId, pFlo->pingPongID));
DC_EXIT_POINT:
DebugExitVOID(FLOPong);
}
//
// FLOPing()
//
void FLOPing
(
PMG_CLIENT pmgClient,
UINT stream,
UINT curtime
)
{
PFLO_STREAM_DATA pStr;
PMG_BUFFER pmgBuffer;
PTSHR_FLO_CONTROL pFlo;
UINT rc;
DebugEntry(FLOPing);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(stream < FLO_MAX_STREAMS);
pStr = pmgClient->flo.pStrData[stream];
ValidateFLOStr(pStr);
rc = MGNewDataBuffer(pmgClient,
MG_TX_PING,
sizeof(TSHR_FLO_CONTROL)+sizeof(MG_INT_PKT_HEADER),
&pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewDataBuffer failed in FLOPing"));
DC_QUIT;
}
//
// Flag ping not needed to avoid serialization problems across the
// sendmessage!
//
pStr->pingNeeded = FALSE;
pFlo = (PTSHR_FLO_CONTROL)pmgBuffer->pDataBuffer;
pmgBuffer->pPktHeader->header.pktLength = TSHR_PKT_FLOW;
//
// Set up ping contents
//
pFlo->packetType = PACKET_TYPE_PING;
pFlo->userID = pmgClient->userIDMCS;
pFlo->stream = (BYTE)stream;
pmgBuffer->channelId = (ChannelID)pStr->channel;
pmgBuffer->priority = (NET_PRIORITY)pStr->priority;
//
// Generate the next ping value to be used
//
pFlo->pingPongID = (BYTE)(pStr->pingValue + 1);
//
// Now decouple the send request
//
TRACE_OUT(("Inserting ping message 0x%08x into pending chain", pmgBuffer));
COM_BasedListInsertBefore(&(pmgClient->pendChain), &(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
//
// Update flow control variables
//
pStr->pingValue = ((pStr->pingValue + 1) & 0xFF);
pStr->lastPingTime = curtime;
pStr->nextPingTime = curtime + pStr->pingTime;
pStr->lastDenialTime = pStr->curDenialTime;
TRACE_OUT(("%u - PING %u sched, next in %u mS (0x%08x:%u)",
pFlo->stream,
pStr->pingValue,
pStr->pingTime,
pStr->channel,
pStr->priority));
DC_EXIT_POINT:
DebugExitVOID(FLOPing);
}
//
// FLOPang()
//
void FLOPang
(
PMG_CLIENT pmgClient,
UINT stream,
UINT userID
)
{
PMG_BUFFER pmgBuffer;
PTSHR_FLO_CONTROL pFlo;
UINT rc;
DebugEntry(FLOPang);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
rc = MGNewDataBuffer(pmgClient,
MG_TX_PANG,
sizeof(TSHR_FLO_CONTROL) + sizeof(MG_INT_PKT_HEADER),
&pmgBuffer);
if (rc != 0)
{
WARNING_OUT(("MGNewDataBuffer failed in FLOPang"));
DC_QUIT;
}
pFlo = (PTSHR_FLO_CONTROL)pmgBuffer->pDataBuffer;
pmgBuffer->pPktHeader->header.pktLength = TSHR_PKT_FLOW;
//
// Set up pang contents
//
pFlo->packetType = PACKET_TYPE_PANG;
pFlo->userID = pmgClient->userIDMCS;
pFlo->stream = (BYTE)stream;
pFlo->pingPongID = 0;
pmgBuffer->channelId = (ChannelID)userID;
pmgBuffer->priority = MG_PRIORITY_HIGHEST;
//
// Now decouple the send request
//
TRACE_OUT(("Inserting pang message 0x%08x into pending chain", pmgBuffer));
COM_BasedListInsertBefore(&(pmgClient->pendChain),
&(pmgBuffer->pendChain));
UT_PostEvent(pmgClient->putTask,
pmgClient->putTask,
NO_DELAY,
NET_MG_SCHEDULE,
0,
0);
DC_EXIT_POINT:
DebugExitVOID(FLOPang);
}
//
// FLOGetStream()
//
UINT FLOGetStream
(
PMG_CLIENT pmgClient,
NET_CHANNEL_ID channel,
UINT priority,
PFLO_STREAM_DATA * ppStr
)
{
UINT i;
UINT cStreams;
DebugEntry(FLOGetStream);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
ASSERT(priority != NET_TOP_PRIORITY);
cStreams = pmgClient->flo.numStreams;
ASSERT(cStreams <= FLO_MAX_STREAMS);
//
// Scan the list of streams for a match.
//
for (i = 0; i < cStreams; i++)
{
//
// Check to ensure that this is a valid stream.
//
if (pmgClient->flo.pStrData[i] == NULL)
{
continue;
}
ValidateFLOStr(pmgClient->flo.pStrData[i]);
//
// If the channel and priority match then we have found the stream.
//
if ((pmgClient->flo.pStrData[i]->channel == channel) &&
(pmgClient->flo.pStrData[i]->priority == priority))
{
break;
}
}
//
// If we hit the end of the list then return FLO_NOT_CONTROLLED.
//
if (i == cStreams)
{
i = FLO_NOT_CONTROLLED;
*ppStr = NULL;
TRACE_OUT(("Uncontrolled stream (0x%08x:%u)",
channel,
priority));
}
else
{
*ppStr = pmgClient->flo.pStrData[i];
TRACE_OUT(("Controlled stream %u (0x%08x:%u)",
i,
channel,
priority));
}
DebugExitDWORD(FLOGetStream, i);
return(i);
}
//
// FUNCTION: FLOAddUser
//
// DESCRIPTION:
//
// Add a new remote user entry for a stream.
//
// PARAMETERS:
//
// userID - ID of the new user (single member channel ID)
// pStr - pointer to the stream to receive the new user.
//
// RETURNS: Nothing
//
//
PFLO_USER FLOAddUser
(
UINT userID,
PFLO_STREAM_DATA pStr
)
{
PFLO_USER pFloUser;
DebugEntry(FLOAddUser);
ValidateFLOStr(pStr);
//
// Allocate memory for the new user entry
//
pFloUser = new FLO_USER;
if (!pFloUser)
{
WARNING_OUT(("FLOAddUser failed; out of memory"));
}
else
{
ZeroMemory(pFloUser, sizeof(*pFloUser));
SET_STAMP(pFloUser, FLOUSER);
//
// Set up the new record
//
TRACE_OUT(("UserID %u - New user, CB = 0x%08x", userID, pFloUser));
pFloUser->userID = (TSHR_UINT16)userID;
//
// Add the new User to the end of the list
//
COM_BasedListInsertBefore(&(pStr->users), &(pFloUser->list));
}
DebugExitVOID(FLOAddUser);
return(pFloUser);
}
//
// FLO_RemoveUser()
//
void FLO_RemoveUser
(
PMG_CLIENT pmgClient,
UINT userID
)
{
PFLO_USER pFloUser;
PBASEDLIST nextUser;
UINT stream;
UINT cStreams;
PFLO_STREAM_DATA pStr;
DebugEntry(FLO_RemoveUser);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
cStreams = pmgClient->flo.numStreams;
ASSERT(cStreams <= FLO_MAX_STREAMS);
//
// Check each stream
//
for (stream = 0; stream < cStreams; stream++)
{
if (pmgClient->flo.pStrData[stream] == NULL)
{
continue;
}
pStr = pmgClient->flo.pStrData[stream];
ValidateFLOStr(pStr);
//
// Remove this user from the queue, if present
//
pFloUser = (PFLO_USER)COM_BasedNextListField(&(pStr->users));
while (&(pFloUser->list) != &(pStr->users))
{
ValidateFLOUser(pFloUser);
//
// Address the follow on record before we free the current
//
nextUser = COM_BasedNextListField(&(pFloUser->list));
//
// Free the current record, if necessary
//
if (pFloUser->userID == userID)
{
//
// Remove from the list
//
TRACE_OUT(("Freeing FLO_USER 0x%08x ID 0x%08x", pFloUser, pFloUser->userID));
COM_BasedListRemove(&(pFloUser->list));
delete pFloUser;
TRACE_OUT(("Stream %u - resetting due to user disappearance",
stream));
ValidateFLOStr(pStr);
pStr->bytesInPipe = 0;
pStr->pingNeeded = TRUE;
pStr->nextPingTime = GetTickCount();
pStr->gotPong = FALSE;
pStr->eventNeeded = FALSE;
break;
}
//
// Move on to the next record in the list
//
pFloUser = (PFLO_USER)nextUser;
}
//
// Now wake the app again for this stream
//
if (pmgClient->flo.callBack != NULL)
{
(*(pmgClient->flo.callBack))(pmgClient,
FLO_WAKEUP,
pStr->priority,
pStr->maxBytesInPipe);
}
}
DebugExitVOID(FLO_RemoveUser);
}
//
// FLOStreamEndControl()
//
void FLOStreamEndControl
(
PMG_CLIENT pmgClient,
UINT stream
)
{
PFLO_USER pFloUser;
PFLO_STREAM_DATA pStr;
PMG_BUFFER pmgBuffer;
DebugEntry(FLOStreamEndControl);
ValidateMGClient(pmgClient);
ASSERT(pmgClient->userAttached);
//
// Convert the stream id into a stream pointer.
//
ASSERT(stream < FLO_MAX_STREAMS);
pStr = pmgClient->flo.pStrData[stream];
ValidateFLOStr(pStr);
//
// Trace out that we are about to end flow control.
//
TRACE_OUT(("Flow control about to end, stream %u, (0x%08x:%u)",
stream,
pStr->channel,
pStr->priority));
//
// First check to see if there are any outstanding buffer CBs with
// pStr set to this stream and reset pStr to null. We need to do this
// as we may then try to dereference pStr when we come to send these
// buffers.
//
pmgBuffer = (PMG_BUFFER)COM_BasedListFirst(&(pmgClient->pendChain),
FIELD_OFFSET(MG_BUFFER, pendChain));
while (pmgBuffer != NULL)
{
ValidateMGBuffer(pmgBuffer);
if (pmgBuffer->type == MG_TX_BUFFER)
{
//
// Set the stream pointer to NULL.
//
pmgBuffer->pStr = NULL;
TRACE_OUT(("Nulling stream pointer in bufferCB: (0x%08x:%u)",
pStr->channel, pStr->priority));
}
pmgBuffer = (PMG_BUFFER)COM_BasedListNext(&(pmgClient->pendChain),
pmgBuffer, FIELD_OFFSET(MG_BUFFER, pendChain));
}
//
// Now free up the list of users.
//
pFloUser = (PFLO_USER)COM_BasedListFirst(&(pStr->users), FIELD_OFFSET(FLO_USER, list));
while (pFloUser != NULL)
{
ValidateFLOUser(pFloUser);
//
// First send the remote user a "pang" to tell them we are not
// interested in their data any more.
//
FLOPang(pmgClient, stream, pFloUser->userID);
//
// Remove the remote user from the list.
//
TRACE_OUT(("Freeing FLO_USER 0x%08x ID 0x%08x", pFloUser, pFloUser->userID));
COM_BasedListRemove(&(pFloUser->list));
delete pFloUser;
//
// Now get the next user in the list.
//
ValidateFLOStr(pStr);
pFloUser = (PFLO_USER)COM_BasedListFirst(&(pStr->users), FIELD_OFFSET(FLO_USER, list));
}
//
// Free the stream data.
//
ASSERT(pStr == pmgClient->flo.pStrData[stream]);
TRACE_OUT(("Freeing FLO_STREAM_DATA 0x%08x", pStr));
delete pStr;
pmgClient->flo.pStrData[stream] = NULL;
//
// Adjust numStreams (if required)
//
if (stream == (pmgClient->flo.numStreams - 1))
{
while ((pmgClient->flo.numStreams > 0) &&
(pmgClient->flo.pStrData[pmgClient->flo.numStreams - 1] == NULL))
{
pmgClient->flo.numStreams--;
}
TRACE_OUT(("numStreams %u", pmgClient->flo.numStreams));
}
DebugExitVOID(FLOStreamEndControl);
}
//
// MGNewCorrelator()
//
// Gets a new correlator for events to a particular MGC client
//
void MGNewCorrelator
(
PMG_CLIENT pmgClient,
WORD * pCorrelator
)
{
ValidateMGClient(pmgClient);
pmgClient->joinNextCorr++;
if (pmgClient->joinNextCorr == 0)
{
pmgClient->joinNextCorr++;
}
*pCorrelator = pmgClient->joinNextCorr;
}