mom_start.c

openPBS的开放源代码

		kill_task(ptask, SIGKILL);
		ptask->ti_qs.ti_exitstat = exiteval;
		ptask->ti_qs.ti_status = TI_STATE_EXITED;
		task_save(ptask);
		sprintf(log_buffer, "task %d terminated", ptask->ti_qs.ti_task);
		LOG_EVENT(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB,
				pjob->ji_qs.ji_jobid, log_buffer);
		exiting_tasks = 1;
	}
}


/*
 * creat the master pty, this particular
 * piece of code depends on multiplexor /dev/ptc
 */

int open_master(rtn_name)
	char **rtn_name;	/* RETURN name of slave pts */
{
	int   ptc;

	if ((ptc = open("/dev/ptc", O_RDWR | O_NOCTTY, 0)) < 0) {
		return (-1);
	}

	*rtn_name  = ttyname(ptc);	/* name of slave side */
	return (ptc);
}




/*
 * struct sig_tbl = map of signal names to numbers,
 * see req_signal() in ../requests.c
 */

struct sig_tbl sig_tbl[] = {
	{ "NULL", 0 },
	{ "HUP", SIGHUP },
	{ "INT", SIGINT },
	{ "QUIT", SIGQUIT },
	{ "ILL", SIGILL },
	{ "TRAP", SIGTRAP },
	{ "ABRT", SIGABRT },
	{ "EMT", SIGEMT },
	{ "FPE", SIGFPE },
	{ "KILL", SIGKILL },
	{ "BUS", SIGBUS },
	{ "SEGV", SIGSEGV },
	{ "SYS", SIGSYS },
	{ "PIPE", SIGPIPE },
	{ "ALRM", SIGALRM },
	{ "TERM", SIGTERM },
	{ "URG", SIGURG },
	{ "STOP", SIGSTOP },
	{ "TSTP", SIGTSTP },
	{ "CONT", SIGCONT },
	{ "CHLD", SIGCHLD },
	{ "TTIN", SIGTTIN },
	{ "TTOU", SIGTTOU },
	{ "IO", SIGIO },
	{ "XCPU", SIGXCPU },
	{ "XFSZ", SIGXFSZ },
	{ "MSG", SIGMSG },
	{ "WINCH", SIGWINCH },
	{ "PWR", SIGPWR },
	{ "USR1", SIGUSR1 },
	{ "USR2", SIGUSR2 },
	{ "PROF", SIGPROF },
	{ "DANGER", SIGDANGER },
	{ "VTALRM", SIGVTALRM },
	{ "MIGRATE", SIGMIGRATE },
	{ "PRE", SIGPRE },
	{ "VIRT", SIGVIRT },
	{ "ALRM1", SIGALRM1 },
	{ "WAITING", SIGWAITING },
	{ "KAP", SIGKAP },
	{ "GRANT", SIGGRANT },
	{ "RETRACT", SIGRETRACT },
	{ "SOUND", SIGSOUND },
	{ "SAK", SIGSAK },
	{ (char *)0, -1 }
};

#if IBM_SP2==2	/* IBM SP2 with PSSP 3.1 */

/*
 * The following routines are used to load and unload the routing information
 * for the IBM high speed switch on the SP.  Each node must specify the same
 * connection information.   At the end of the job the information is unloaded.
 */

static int name_to_sw_num(nn)
	char *nn;
{
	int   i;
	char *pa;
	char *pb;
	extern struct swtbl_num swtbl_num[];
	extern int    ibm_sp2_num_nodes;

	for (i=0; i<ibm_sp2_num_nodes; i++) {
		pa = nn;
		pb = swtbl_num[i].sw_name;
		while (*pa && *pb && (*pa == *pb)) {
			pa++;
			pb++;
		}
		if ( ((*pa == '\0') && (*pb == '\0')) ||
		     ((*pa == '\0') && (*pb == '.'))  ||
		     ((*pa == '.')  && (*pb == '\0')) )
			return swtbl_num[i].sw_num;
	}
	return -1;
}
	
	
static int build_swtbl_array(pjob, psa)
	job		     *pjob;
	struct ST_NODE_INFO **psa;
{
	int			 i;
	int			 j;
	struct ST_NODE_INFO *swtbl;

	swtbl = (struct ST_NODE_INFO *)calloc(pjob->ji_numvnod, 
			sizeof(struct ST_NODE_INFO));
	if (swtbl == NULL)
		return PBSE_SYSTEM;

	for (i=0; i<pjob->ji_numvnod; i++) {
		/* pull node name out of vnodent/hnodent struct */
		(void)strcpy((swtbl+i)->st_node_name, 
				pjob->ji_vnods[i].vn_host->hn_host);

		/* now find switch node number that matches name */

		if ((j = name_to_sw_num((swtbl+i)->st_node_name)) == -1) {
			(void)free(swtbl);
			return PBSE_UNKNODE;
		}

		(swtbl+i)->st_virtual_task_id = i;
		(swtbl+i)->st_switch_node_num = j;
		(swtbl+i)->st_window_id = pjob->ji_vnods[i].vn_index;
	}

	*psa = swtbl;
	return PBSE_NONE;
}

/*
 * load_sp_switch - load the IBM SP switch table with the nodes/window_id
 *	to be used by this job.  Also write a file into the "aux" directory
 *	with the array of window_ids in task order.  This file is used by
 *	pbspd.c to place the correct window_id into the environment based on
 *	the value of MP_CHILD passed by IBM's poe.
 */

int load_sp_switch(pjob)
	job	*pjob;
{
	int		     i;
	int		     rc = 0;
	struct ST_NODE_INFO *pswa;
	static char *id = "load_sp_switch";
	char buf[MAXPATHLEN+1];
	FILE *fwin;
	extern int internal_state;

	
	sscanf(pjob->ji_wattr[(int)JOB_ATR_Cookie].at_val.at_str,"%x",&job_key);

	if (job_key < 0)
		job_key = -job_key;
	if ((rc = build_swtbl_array(pjob, &pswa)) != PBSE_NONE) {
		sprintf(log_buffer, "build swtbl node array failed %d", rc);
		log_record(PBSEVENT_DEBUG, 0, id, log_buffer);
		return -1;
	}

	rc = swtbl_load_table(ST_VERSION, pjob->ji_qs.ji_un.ji_momt.ji_exuid,
			      getpid(), job_key, mom_host, pjob->ji_numvnod,
			      pjob->ji_qs.ji_jobid, pswa);
	if (rc != ST_SUCCESS) {
		sprintf(log_buffer, "swtbl_load_table failed with %d", rc);
		log_record(PBSEVENT_SYSTEM|PBSEVENT_ADMIN|PBSEVENT_JOB,
			   PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid,log_buffer);
		internal_state |= INUSE_DOWN;
		state_to_server(1);	/* tell server we are down */
	} else {

	    /* success */
	    (void)sprintf(buf, "%s/aux/%s.SW", path_home, pjob->ji_qs.ji_jobid);
	    if ((fwin = fopen(buf, "w")) == NULL) {
		sprintf(log_buffer, "cannot open %s", buf);
		log_err(errno, id, log_buffer);
		return -1;
	    }
	    (void)fchmod(fileno(fwin), 0644);
	    for (i=0; i<pjob->ji_numvnod; i++)
		fprintf(fwin, "%d\n", pjob->ji_vnods[i].vn_index);
	    (void)fclose(fwin);
	}
	(void)free(pswa);
	return rc;
}

/*
 * unload_sp_switch - unload the job/node information from the switch
 *	Also remove the aux *.SW file created when the switch is loaded above.
 */

void unload_sp_switch(pjob)
	job	*pjob;
{
	int		     i;
	int		     rc = 0;
	char		     buf[MAXPATHLEN+1];
	struct vnodent	    *pvp;
	extern int 	     internal_state;

	for (i=0; i<pjob->ji_numvnod; ++i) {
	    pvp = &pjob->ji_vnods[i];
	    if (pvp->vn_host->hn_node == pjob->ji_nodeid) {
		rc = swtbl_unload_table(ST_VERSION, "css0",
				    pjob->ji_qs.ji_un.ji_momt.ji_exuid, 
				    pvp->vn_index);
		DBPRT(("Unloading switch window %d\n", pvp->vn_index))

		if (rc != ST_SUCCESS) {
		    sprintf(log_buffer,"error %d unloading switch table window %d for job %s", rc, pvp->vn_index, pjob->ji_qs.ji_jobid);
		    log_err(PBSE_SYSTEM, "unload_sp_switch", log_buffer);

		    if (rc != ST_SWITCH_NOT_LOADED) {
			rc = swtbl_clean_table(ST_VERSION, "css0",
					       ST_ALWAYS_KILL,pvp->vn_index);
		    } else if (rc != ST_SUCCESS) {
			sprintf(log_buffer,"error %d cleaning switch table window %d for job %s", rc, pvp->vn_index, pjob->ji_qs.ji_jobid);
			log_err(PBSE_SYSTEM, "unload_sp_switch", log_buffer);
			internal_state |= INUSE_DOWN;
			state_to_server(1);	/* tell server we are down */
		    }
		}
	    }
	}

	(void)sprintf(buf, "%s/aux/%s.SW", path_home, pjob->ji_qs.ji_jobid);
	(void)unlink(buf);
	return;
} 

/* 
 * query_adp - query the SP switch adaptor, are we on line
 */

void query_adp()
{
	int  rc;
	enum ST_ADAPTER_STATUS st;
	extern int internal_state;
	static char *id = "query_adp";

	if ((rc = swtbl_query_adapter(ST_VERSION, "css0", &st)) != ST_SUCCESS) {
		if ((internal_state & INUSE_DOWN) == 0) {
			log_record(PBSEVENT_SYSTEM, rc, id, 
							"cannot query adaptor");
			internal_state |= INUSE_DOWN|UPDATE_MOM_STATE;
		}
		return;
	}
	switch (st) {
	    case ADAPTER_READY:
		if (internal_state & INUSE_DOWN) {
			internal_state &= ~INUSE_DOWN;	/* we are not down */
			internal_state |=  UPDATE_MOM_STATE;
			log_record(PBSEVENT_SYSTEM, 0, id, "adaptor up");
		}
		break;
		
	    case ADAPTER_NOTREADY:
		if ((internal_state & INUSE_DOWN) == 0) {
			/* mark that we are down */
			internal_state |= INUSE_DOWN|UPDATE_MOM_STATE;
			log_record(PBSEVENT_SYSTEM, 0, id, "adaptor down");
		}
		break;
	}
}
	
#endif	/* IBM SP */