/* distGapLib.c - distributed group agreement protocol library *//* Copyright 1999 Wind River Systems, Inc. */#include "copyright_wrs.h"/*modification history--------------------01i,24may99,drm  added vxfusion prefix to VxFusion related includes01h,11feb99,drm  fixed endian bugs in GAP protocol - added htons() / ntohs()01g,11sep98,drm  changed #include "distStatLib.h" to "private/distLibP.h"01f,20may98,drm  removed local variable distNodeLclState which was no longer                 being used.01e,13may98,ur   review of some comments01d,15apr98,ur   replaced misleading comments01c,13mar98,ur   fixed crash detection01b,07jan98,ur   fixed "group id already in use"01a,15jul97,ur   written*//*DESCRIPTIONThis library contains the "group agreement protocol" (GAP).*/#include "vxWorks.h"#if defined (DIST_GAP_REPORT) || defined (DIST_DIAGNOSTIC)#include "stdio.h"#endif#include "stdlib.h"#include "string.h"#include "sllLib.h"#include "dllLib.h"#include "msgQLib.h"#include "taskLib.h"#include "netinet/in.h"#include "private/semLibP.h"#include "vxfusion/msgQDistGrpLib.h"#include "vxfusion/distIfLib.h"#include "vxfusion/private/distLibP.h"#include "vxfusion/private/distNameLibP.h"#include "vxfusion/private/msgQDistGrpLibP.h"#include "vxfusion/private/distNetLibP.h"#include "vxfusion/private/distNodeLibP.h"LOCAL DL_LIST			gapInProgress;LOCAL SEMAPHORE			distGapSemaphore;/* local prototypes */LOCAL void			distGapCatchCrashed (DIST_NODE_ID nodeId);LOCAL void			distGapPh1Done (DIST_GRP_DB_NODE *pDistGrp);LOCAL STATUS		distGapTry (DIST_NODE_ID distNodeId,						DIST_GRP_DB_NODE *pDbNode);LOCAL BOOL			distGapOutstandConc (DIST_NODE_DB_NODE *pDistNodeDbNode,						SL_LIST *pOutstand);LOCAL BOOL			distGapOutstandOk (DIST_GAP_NODE *pDistGapNode,						DIST_NODE_ID distNodeIdOk);LOCAL void			distGapOutstandDel (DIST_GAP_NODE *pDistGapNode,						DIST_GAP_RESPONSE *pDistGapResponse,						DIST_GAP_RESPONSE *pDistGapResponsePrev);LOCAL DIST_STATUS	distGapInput (DIST_NODE_ID nodeIdSrc,						DIST_TBUF_HDR *pTBufHdr);#ifdef DIST_GAP_TASKLOCAL void			distGapTimerTask (void);LOCAL void			distGapTimer (void);#endif/********************************************************************************* distGapLibInit - ** NOMANUAL*/STATUS distGapLibInit()	{#ifdef DIST_GAP_TASK	int tid;#endif	dllInit (&gapInProgress);#ifdef DIST_GAP_TASK	tid = taskSpawn ("tDistGAP",			DIST_GAP_MGR_PRIO,			VX_SUPERVISOR_MODE | VX_UNBREAKABLE,			DIST_GAP_MGR_STACK_SZ,			(FUNCPTR) distGapTimerTask,			0, 0, 0, 0, 0, 0, 0, 0, 0, 0);				if (tid == ERROR)		return (ERROR);#endif	distCtl (DIST_CTL_CRASHED_HOOK, (int) distGapCatchCrashed);	/*	 * Add GAP input function to table of services.	 */	return (distNetServAdd (DIST_PKT_TYPE_GAP, distGapInput,			DIST_GAP_SERV_NAME, DIST_GAP_SERV_NET_PRIO,			DIST_GAP_SERV_TASK_PRIO, DIST_GAP_SERV_TASK_STACK_SZ));	}/********************************************************************************* distGapCatchCrashed - get notified by a node crash** NOMANUAL*/LOCAL void distGapCatchCrashed	(	DIST_NODE_ID crashedId	)	{	DIST_GRP_DB_NODE	*pGrpDone;	DIST_GAP_NODE		*pGapNode;	BOOL				more;#ifdef DIST_GAP_REPORT	printf ("distGapCatchCrashed: node %ld crashed\n", crashedId);#endif	pGapNode = (DIST_GAP_NODE *) DLL_FIRST (&gapInProgress);	while (pGapNode != NULL)		{		msgQDistGrpDbLock();		pGrpDone = pGapNode->pGapGrp;		distGapLock();		more = distGapOutstandOk (pGapNode, crashedId);		distGapUnlock();		pGapNode = (DIST_GAP_NODE *) DLL_NEXT (pGapNode);		if (more == FALSE)	/* this was the last node we waited for */			{			msgQDistGrpLclSetState (pGrpDone, DIST_GRP_STATE_GLOBAL);			msgQDistGrpLclSetCreator (pGrpDone, distNodeLocalGetId());			msgQDistGrpDbUnlock();			distGapPh1Done (pGrpDone);			}		else			msgQDistGrpDbUnlock();		}	}/********************************************************************************* distGapOutstandLink - link all operational nodes to GAP structure** NOMANUAL*/void distGapOutstandLink	(	DIST_GAP_NODE	*distGapNode	)	{	sllInit (&distGapNode->gapOutstand);	distNodeEach (distGapOutstandConc, (int) &distGapNode->gapOutstand);	}/********************************************************************************* distGapOutstandConc - called by distGapOutstandLink()** NOMANUAL*/LOCAL BOOL distGapOutstandConc	(	DIST_NODE_DB_NODE	*pDistNodeDbNode,	SL_LIST				*pOutstand	)	{	DIST_GAP_RESPONSE	*pDistGapResp;	if (pDistNodeDbNode->nodeState == DIST_NODE_STATE_OPERATIONAL &&		pDistNodeDbNode->nodeId != DIST_IF_BROADCAST_ADDR)		{		pDistGapResp =				(DIST_GAP_RESPONSE *) malloc (sizeof (DIST_GAP_RESPONSE));		if (pDistGapResp == NULL)			return (ERROR);		pDistGapResp->pGapResponseNode = pDistNodeDbNode;		sllPutAtHead (pOutstand, (SL_NODE *) pDistGapResp);		}	return (TRUE);	/* continue */	}/********************************************************************************* distGapOutstandUnlink - Free all nodes linked to the GAP structure** NOMANUAL*/void distGapOutstandUnlink	(	DIST_GAP_NODE	*distGapNode	)	{	DIST_GAP_RESPONSE	*pDistGapResponse;	DIST_GAP_RESPONSE	*pDistGapResponsePrev = NULL;	if (distGapNode == NULL)		return;	pDistGapResponse = (DIST_GAP_RESPONSE *)			SLL_FIRST (&distGapNode->gapOutstand);	while (pDistGapResponse != NULL)		{		sllRemove (&distGapNode->gapOutstand, (SL_NODE *) pDistGapResponse,				(SL_NODE *) pDistGapResponsePrev);		pDistGapResponsePrev = pDistGapResponse;		pDistGapResponse = (DIST_GAP_RESPONSE *) SLL_NEXT (pDistGapResponse);		free (pDistGapResponsePrev);		}	}/********************************************************************************* distGapOutstandOk - Delete a single node from the GAP structure's node list** RETURNS: TRUE if there are more outstanding responses, or FALSE if the* job is done.** NOMANUAL*/LOCAL BOOL distGapOutstandOk	(	DIST_GAP_NODE	*pDistGapNode,	DIST_NODE_ID	distNodeIdOk	)	{	DIST_GAP_RESPONSE	*pDistGapResp;	DIST_GAP_RESPONSE	*pDistGapRespPrev;	pDistGapResp = (DIST_GAP_RESPONSE *)			SLL_FIRST (&pDistGapNode->gapOutstand);	pDistGapRespPrev = NULL;	while (pDistGapResp != NULL &&		   pDistGapResp->pGapResponseNode->nodeId != distNodeIdOk)		{		pDistGapRespPrev = pDistGapResp;		pDistGapResp = (DIST_GAP_RESPONSE *) SLL_NEXT (pDistGapResp);		}	if (pDistGapResp != NULL)		{		distGapOutstandDel (pDistGapNode, pDistGapResp, pDistGapRespPrev);		if (SLL_FIRST (&pDistGapNode->gapOutstand) == NULL)			return (FALSE);		}	return (TRUE);	}/********************************************************************************* distGapOutstandDel - Delete a single node ** NOMANUAL*/LOCAL void distGapOutstandDel	(	DIST_GAP_NODE		*pDistGapNode,	DIST_GAP_RESPONSE	*pDistGapResponse,	DIST_GAP_RESPONSE	*pDistGapResponsePrev	)	{	sllRemove ((SL_LIST *) &pDistGapNode->gapOutstand,			(SL_NODE *) pDistGapResponse, (SL_NODE *) pDistGapResponsePrev);	free (pDistGapResponse);#ifdef DIST_GAP_REPORT	{	SL_NODE	*pNode = SLL_FIRST (&pDistGapNode->gapOutstand);	int		i;	for (i=0; pNode; i++)		pNode = SLL_NEXT (pNode);	printf ("distGapOutstandDel: %d nodes outstanding\n", i);	}#endif	}/********************************************************************************* distGapPh1Done - phase 1 is done** NOMANUAL*/LOCAL void distGapPh1Done	(	DIST_GRP_DB_NODE	*pDistGrp	)	{	/*	 * List of outstanding responses is empty now. We are done.	 * Send SET telegram and wakeup sleeping task.	 */	DIST_PKT_GAP_SET	distPktGapSet;#ifdef DIST_GAP_REPORT	printf ("distGapInput/OK: GAP phase 1 done.\n");#endif	/*	 * Build SET telegram and broadcast it.	 */	distPktGapSet.setHdr.pktType = DIST_PKT_TYPE_GAP;	distPktGapSet.setHdr.pktSubType = DIST_PKT_TYPE_GAP_SET;	distPktGapSet.setId = htons (pDistGrp->grpDbId);	strcpy ((char *) &distPktGapSet.setName,			(char *) &pDistGrp->grpDbName);#ifdef DIST_GAP_REPORT	printf ("distGapInput/OK: GAP phase 2 -- broadcast SET `%s'/%d.\n",			distPktGapSet.setName, distPktGapSet.setId);#endif	distNetSend (DIST_IF_BROADCAST_ADDR, (DIST_PKT *) &distPktGapSet,			sizeof (distPktGapSet), WAIT_FOREVER, DIST_GAP_PRIO);	/*	 * Wakeup initiator.	 */	semGive (&pDistGrp->pGrpDbGapNode->gapWaitFor);	}/********************************************************************************* distGapNodeInit - Initialize a GAP node** NOMANUAL*/void distGapNodeInit	(	DIST_GAP_NODE		*pDistGapNode,	DIST_GRP_DB_NODE	*pDistGrpDbNode,	BOOL				link	)	{	if (link)		distGapOutstandLink(pDistGapNode);	semBInit (&(pDistGapNode->gapWaitFor), SEM_Q_FIFO, SEM_EMPTY);	pDistGapNode->pGapGrp		= pDistGrpDbNode;	pDistGapNode->gapTimeout	= DIST_GAP_TRY_TIMO;	pDistGapNode->gapRetries	= 1;	distGapLock();	dllInsert (&gapInProgress, NULL, &(pDistGapNode->gapLink));	distGapUnlock();	}/********************************************************************************* distGapNodeDelete - Delete a GAP node** NOMANUAL*/void distGapNodeDelete	(	DIST_GAP_NODE	*pDistGapNode	)	{	distGapLock();	dllRemove (&gapInProgress, (DL_NODE *) &(pDistGapNode->gapLink));	distGapUnlock();	}/********************************************************************************* distGapStart - Start GAP (group agreement protocol)** Do a TRY in order to create a new group.** NOMANUAL*/DIST_MSG_Q_GRP_ID distGapStart	(	DIST_GRP_DB_NODE	*pDistGrpDbNode	)	{	DIST_GAP_NODE		distGapNode;	DIST_MSG_Q_GRP_ID	uniqGrpId;	if (distNodeGetNumNodes(DIST_NODE_NUM_NODES_ALIVE) == 1)		{		/* we are alone */		msgQDistGrpLclSetState (pDistGrpDbNode, DIST_GRP_STATE_GLOBAL);		msgQDistGrpLclSetCreator (pDistGrpDbNode, distNodeLocalGetId());		return (pDistGrpDbNode->grpDbId);		}	/*	 * Init GAP Node and link it to the group database node and	 * the list of active GAPs.	 * GAP initialization includes the linking of all nodes,	 * currently known in the system. When a response comes in,	 * the linked list is updated (the node is removed from the	 * list).	 */	distGapNodeInit(&distGapNode, pDistGrpDbNode, TRUE);	pDistGrpDbNode->pGrpDbGapNode = &distGapNode;	/*	 * Send TRY.	 */	distGapTry (DIST_IF_BROADCAST_ADDR, pDistGrpDbNode);	/*	 * Wait on GAP Node.	 */	semTake (&distGapNode.gapWaitFor, WAIT_FOREVER);	/*	 * Cleanup	 */	if (pDistGrpDbNode->pGrpDbGapNode)		{		DIST_GAP_NODE	*pGapNode = pDistGrpDbNode->pGrpDbGapNode;		pDistGrpDbNode->pGrpDbGapNode = NULL;		distGapNodeDelete (pGapNode);		}#ifdef DIST_DIAGNOSTIC	if (pDistGrpDbNode->grpDbState != DIST_GRP_STATE_GLOBAL)		distLog ("distGapStart: group has no global state\n");#endif	/*	 * Ask for group id.	 */	msgQDistGrpDbLock();	uniqGrpId = msgQDistGrpLclGetId (pDistGrpDbNode);	msgQDistGrpDbUnlock();	return (uniqGrpId);	}/********************************************************************************* distGapTry - Send a TRY, for a specified group** NOMANUAL*/LOCAL STATUS distGapTry	(	DIST_NODE_ID		distNodeId,		/* node address */	DIST_GRP_DB_NODE	*pDistGrpDbNode	/* group to try */	)	{	DIST_PKT_GAP_TRY	distPktGapTry;	STATUS				status;	/*	 * Build TRY telegram and send it.	 */