Esempio n. 1
0
/**
 *
 * @brief
 * 		Send a job over the network to some other server or MOM.
 * @par
 * 		Under Linux/Unix, this starts a child process to do the work.
 *		Connect to the destination host and port,
 * 		and go through the protocol to transfer the job.
 * 		Signals are blocked.
 *
 * @param[in]	jobp	-	pointer to the job being sent.
 * @param[in]	hostaddr	-	the address of host to send job to, host byte order.
 * @param[in]	port	-	the destination port, host byte order
 * @param[in]	move_type	-	the type of move (e.g. MOVE_TYPE_exec)
 * @param[in]	post_func	-	the function to execute once the child process
 *								sending job completes (Linux/Unix only)
 * @param[in]	data	-	input data to 'post_func'
 *
 * @return	int
 * @retval	2	parent	: success (child forked)
 * @retval	-1	parent	: on failure (pbs_errno set to error number)
 * @retval	SEND_JOB_OK	child	: 0 success, job sent
 * @retval	SEND_JOB_FATAL	child	: 1 permenent failure or rejection,
 * @retval	SEND_JOB_RETRY	child	: 2 failed but try again
 * @retval	SEND_JOB_NODEDW child	: 3 execution node down, retry different node
 */
int
send_job(job *jobp, pbs_net_t hostaddr, int port, int move_type,
	void (*post_func)(struct work_task *), struct batch_request *preq)
{

#ifdef WIN32
	char	cmdline[80];
	pio_handles	pio;
	char	buf[4096];
	struct work_task *ptask;
	int	newstate;
	int	newsub;
	long	tempval;
	char	script_name[MAXPATHLEN+1];
	int 		gridproxy_cred = 0;

#ifdef  PBS_CRED_GRIDPROXY
	if (jobp->ji_extended.ji_ext.ji_credtype == PBS_CREDTYPE_GRIDPROXY)
		gridproxy_cred = 1;
#endif

	if (pbs_conf.pbs_use_tcp == 1 && move_type == MOVE_TYPE_Exec && gridproxy_cred == 0) {
		return (send_job_exec(jobp, hostaddr, port, preq));
	}

	sprintf(cmdline, "%s/sbin/pbs_send_job", pbs_conf.pbs_exec_path);

	if (win_popen(cmdline, "w", &pio, NULL) == 0) {
		errno = GetLastError();
		pbs_errno = errno;
		(void)sprintf(log_buffer, "executing %s for job %s failed errno=%d", cmdline, jobp->ji_qs.ji_jobid, errno);
		log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB, LOG_ERR,
			jobp->ji_qs.ji_jobid, log_buffer);
		/* force re-eval of job state out of Transit */
		svr_evaljobstate(jobp, &newstate, &newsub, 1);
		svr_setjobstate(jobp, newstate, newsub);

		win_pclose(&pio);
		return (-1);
	}

	ptask = set_task(WORK_Deferred_Child, (long)pio.pi.hProcess, post_func, preq);
	if (!ptask) {
		log_err(errno, __func__, msg_err_malloc);
		errno = ENOMEM;
		pbs_errno = errno;
		win_pclose(&pio);
		/* force re-eval of job state out of Transit */
		svr_evaljobstate(jobp, &newstate, &newsub, 1);
		svr_setjobstate(jobp, newstate, newsub);
		return (-1);
	} else {
		ptask->wt_parm2 = jobp;
		append_link(&((job *)jobp)->ji_svrtask, &ptask->wt_linkobj, ptask);
	}

	script_name[0] = '\0';
	/* if job has a script read it from database */
	if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
		/*
		 * copy the job script from database to a temp file
		 * PBSD_jscript works with a file
		 * delete it at the end of the send
		 */
		if (svr_create_tmp_jobscript(jobp, &script_name) != 0) {
			pbs_errno = PBSE_SYSTEM;
			snprintf(log_buffer, sizeof(log_buffer),
				"Failed to create temporary job script for job %s",
				jobp->ji_qs.ji_jobid);
			log_err(pbs_errno, "send_job", log_buffer);
			win_pclose2(&pio);
			return (-1);
		}
	}

	addpid(pio.pi.hProcess);

	/* our job is to calc eligible time accurately and save it */
	/* on new server, accrue type should be calc afresh */
	/* Note: if job is being sent for execution on mom, then don't calc eligible time */

	if ((jobp->ji_wattr[(int)JOB_ATR_accrue_type].at_val.at_long == JOB_ELIGIBLE) &&
		(server.sv_attr[(int)SRV_ATR_EligibleTimeEnable].at_val.at_long == 1) &&
		(move_type != MOVE_TYPE_Exec)) {
		tempval = ((long)time_now - jobp->ji_wattr[(int)JOB_ATR_sample_starttime].at_val.at_long);
		jobp->ji_wattr[(int)JOB_ATR_eligible_time].at_val.at_long += tempval;
		jobp->ji_wattr[(int)JOB_ATR_eligible_time].at_flags |= ATR_VFLAG_MODCACHE;
	}

	/* in windows code, a child process "w32_send_job" handles the send
	 * This needs the job information, so we save using the filesystem
	 * This avoids the child process from having to "connect" to the database again
	 * The file is deleted by the send_job child process when it has done recovering the job
	 */
	job_save_fs(jobp, SAVEJOB_FULLFORCE);	/* so the spawned process can get a fresh copy of job */

	if (*jobp->ji_qs.ji_fileprefix != '\0')
		sprintf(buf, "jobfile=%s%s\n", jobp->ji_qs.ji_fileprefix, JOB_FILE_SUFFIX);
	else
		sprintf(buf, "jobfile=%s%s\n", jobp->ji_qs.ji_jobid, JOB_FILE_SUFFIX);

	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "destaddr=%ld\n", hostaddr);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "destport=%d\n", port);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "move_type=%d\n", move_type);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "in_server=%d\n", is_linked(&svr_alljobs, &jobp->ji_alljobs));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "server_name=%s\n", (server_name?server_name:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "server_host=%s\n", (server_host?server_host:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "server_addr=%ld\n", pbs_server_addr);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "server_port=%d\n", pbs_server_port_dis);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "log_file=%s\n", (log_file?log_file:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_log=%s\n", (path_log?path_log:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_jobs=%s\n", (path_jobs?path_jobs:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_spool=%s\n", (path_spool?path_spool:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_rescdef=%s\n", (path_rescdef?path_rescdef:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_users=%s\n", (path_users?path_users:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "path_hooks_workdir=%s\n",
		(path_hooks_workdir?path_hooks_workdir:""));
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "svr_history_enable=%ld\n", svr_history_enable);
	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "svr_history_duration=%ld\n", svr_history_duration);
	win_pwrite(&pio, buf, strlen(buf));

	if ( (server.sv_attr[SRV_ATR_ssignon_enable].at_flags & \
                                                ATR_VFLAG_SET) && \
             (server.sv_attr[SRV_ATR_ssignon_enable].at_val.at_long == 1) )
		strcpy(buf, "single_signon_password_enable=1\n");
	else
		strcpy(buf, "single_signon_password_enable=0\n");

	win_pwrite(&pio, buf, strlen(buf));

	sprintf(buf, "script_name=%s\n", script_name);
	win_pwrite(&pio, buf, strlen(buf));

	strcpy(buf, "quit\n");
	win_pwrite(&pio, buf, strlen(buf));
	win_pclose2(&pio);	/* closes all handles except the process handle */
	return (2);
#else
	pbs_list_head	 attrl;
	enum conn_type   cntype = ToServerDIS;
	int		 con;
	char		*credbuf = NULL;
	size_t		 credlen = 0;
	char		*destin = jobp->ji_qs.ji_destin;
	int		 encode_type;
	int		 i;
	char		 job_id[PBS_MAXSVRJOBID+1];
	attribute	*pattr;
	pid_t		 pid;
	struct attropl  *pqjatr;      /* list (single) of attropl for quejob */
	char		 script_name[MAXPATHLEN+1];
	struct work_task *ptask;
	struct  hostent *hp;
	struct in_addr   addr;
	long		 tempval;
	int 		gridproxy_cred = 0;
	int 		rpp = 0;

#ifdef  PBS_CRED_GRIDPROXY
	if (jobp->ji_extended.ji_ext.ji_credtype == PBS_CREDTYPE_GRIDPROXY)
		gridproxy_cred = 1;
#endif

	if (pbs_conf.pbs_use_tcp == 1 && move_type == MOVE_TYPE_Exec && gridproxy_cred == 0) {
		return (send_job_exec(jobp, hostaddr, port, preq));
	}

	script_name[0] = '\0';
	/* if job has a script read it from database */
	if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
		/*
		 * copy the job script from database to a temp file
		 * PBSD_jscript works with a file
		 * delete it at the end of the send
		 */
		if (svr_create_tmp_jobscript(jobp, script_name) != 0) {
			pbs_errno = PBSE_SYSTEM;
			snprintf(log_buffer, sizeof(log_buffer),
				"Failed to create temporary job script for job %s",
				jobp->ji_qs.ji_jobid);
			log_err(pbs_errno, "send_job", log_buffer);
			return -1;
		}
	}

	pid = fork();
	if (pid == -1) {	/* Error on fork */
		log_err(errno, __func__, "fork failed\n");
		pbs_errno = PBSE_SYSTEM;
		return -1;
	}

	if (pid != 0) {		/* The parent (main server) */

		ptask = set_task(WORK_Deferred_Child, pid, post_func, preq);
		if (!ptask) {
			log_err(errno, __func__, msg_err_malloc);
			return (-1);
		} else {
			ptask->wt_parm2 = jobp;
			append_link(&((job *)jobp)->ji_svrtask,
				&ptask->wt_linkobj, ptask);
		}
		return 2;
	}

	/*
	 * the child process
	 *
	 * set up signal cather for error return
	 */
	DBPRT(("%s: child started, sending to port %d\n", __func__, port))
	rpp_terminate();

	/* Unprotect child from being killed by kernel */
	daemon_protect(0, PBS_DAEMON_PROTECT_OFF);

#ifdef WIN32
	/* get host name */
	/*
	 * If host address is loopback address then do not resolve with dns
	 * Use "localhost" as the host name.
	 */
	if ((htonl(hostaddr) == loopback_addr->sin_addr.s_addr)) {
		(void)get_credential(LOCALHOST_SHORTNAME, jobp, PBS_GC_BATREQ,
			&credbuf, &credlen);
	} else {
#endif
		addr.s_addr = htonl(hostaddr);
		hp = gethostbyaddr((void *)&addr, sizeof(struct in_addr), AF_INET);
		if (hp == NULL) {
			sprintf(log_buffer, "%s: h_errno=%d",
				inet_ntoa(addr), h_errno);
			log_err(-1, __func__, log_buffer);
		} else {
			/* read any credential file */
			(void)get_credential(hp->h_name, jobp, PBS_GC_BATREQ,
				&credbuf, &credlen);
		}
#ifdef WIN32
	}
#endif

	/* encode job attributes to be moved */

	CLEAR_HEAD(attrl);

	/* select attributes/resources to send based on move type */

	if (move_type == MOVE_TYPE_Exec) {
		resc_access_perm = ATR_DFLAG_MOM;
		encode_type = ATR_ENCODE_MOM;
		cntype = ToServerDIS;
	} else {
		resc_access_perm = ATR_DFLAG_USWR | ATR_DFLAG_OPWR |
			ATR_DFLAG_MGWR | ATR_DFLAG_SvRD;
		encode_type = ATR_ENCODE_SVR;
		svr_dequejob(jobp);	/* clears default resource settings */
	}

	/* our job is to calc eligible time accurately and save it */
	/* on new server, accrue type should be calc afresh */
	/* Note: if job is being sent for execution on mom, then don't calc eligible time */

	if ((jobp->ji_wattr[(int)JOB_ATR_accrue_type].at_val.at_long == JOB_ELIGIBLE) &&
		(server.sv_attr[(int)SRV_ATR_EligibleTimeEnable].at_val.at_long == 1) &&
		(move_type != MOVE_TYPE_Exec)) {
		tempval = ((long)time_now - jobp->ji_wattr[(int)JOB_ATR_sample_starttime].at_val.at_long);
		jobp->ji_wattr[(int)JOB_ATR_eligible_time].at_val.at_long += tempval;
		jobp->ji_wattr[(int)JOB_ATR_eligible_time].at_flags |= ATR_VFLAG_MODCACHE;
	}

	pattr = jobp->ji_wattr;
	for (i=0; i < (int)JOB_ATR_LAST; i++) {
		if ((job_attr_def+i)->at_flags & resc_access_perm) {
			(void)(job_attr_def+i)->at_encode(pattr+i, &attrl,
				(job_attr_def+i)->at_name, (char *)0,
				encode_type, NULL);
		}
	}
	attrl_fixlink(&attrl);


	/* save the job id for when after we purge the job */

	(void)strcpy(job_id, jobp->ji_qs.ji_jobid);

	pbs_errno = 0;
	con = -1;

	for (i=0; i<RETRY; i++) {

		/* connect to receiving server with retries */

		if (i > 0) {	/* recycle after an error */
			if (con >= 0)
				svr_disconnect(con);
			if (should_retry_route(pbs_errno) == -1) {
				/* delete the temp script file */
				unlink(script_name);
				exit(SEND_JOB_FATAL);	/* fatal error, don't retry */
			}
			sleep(1<<i);
		}
		if ((con = svr_connect(hostaddr, port, 0, cntype, rpp)) ==
			PBS_NET_RC_FATAL) {
			(void)sprintf(log_buffer, "send_job failed to %lx port %d",
				hostaddr, port);
			log_err(pbs_errno, __func__, log_buffer);

			/* delete the temp script file */
			unlink(script_name);

			if ((move_type == MOVE_TYPE_Exec) && (pbs_errno == PBSE_BADCRED))
				exit(SEND_JOB_NODEDW);

			exit(SEND_JOB_FATAL);
		} else if (con == PBS_NET_RC_RETRY) {
			pbs_errno = ECONNREFUSED;	/* should retry */
			continue;
		}

		/*
		 * if the job is substate JOB_SUBSTATE_TRNOUTCM which means
		 * we are recovering after being down or a late failure, we
		 * just want to send the commit"
		 */

		if (jobp->ji_qs.ji_substate != JOB_SUBSTATE_TRNOUTCM) {

			if (jobp->ji_qs.ji_substate != JOB_SUBSTATE_TRNOUT) {
				jobp->ji_qs.ji_substate = JOB_SUBSTATE_TRNOUT;
			}

			pqjatr = &((svrattrl *)GET_NEXT(attrl))->al_atopl;
			if (PBSD_queuejob(con, jobp->ji_qs.ji_jobid, destin,
				pqjatr, (char *)0, rpp, NULL) == 0) {
				if (pbs_errno == PBSE_JOBEXIST &&
					move_type == MOVE_TYPE_Exec) {
					/* already running, mark it so */
					log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_JOB,
						LOG_INFO, jobp->ji_qs.ji_jobid,
						"Mom reports job already running");
					exit(SEND_JOB_OK);
				}
				else if ((pbs_errno == PBSE_HOOKERROR) ||
					(pbs_errno == PBSE_HOOK_REJECT)  ||
					(pbs_errno == PBSE_HOOK_REJECT_RERUNJOB)  ||
					(pbs_errno == PBSE_HOOK_REJECT_DELETEJOB)) {
					char		name_buf[MAXPATHLEN+1];
					int		rfd;
					int		len;
					char		*reject_msg;
					int		err;

					err = pbs_errno;

					reject_msg = pbs_geterrmsg(con);
					(void)sprintf(log_buffer,
						"send of job to %s failed error = %d reject_msg=%s",
						destin, err,
						reject_msg?reject_msg:"");
					log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB,
						LOG_INFO, jobp->ji_qs.ji_jobid,
						log_buffer);

					(void)strcpy(name_buf, path_hooks_workdir);
					(void)strcat(name_buf, jobp->ji_qs.ji_jobid);
					(void)strcat(name_buf, HOOK_REJECT_SUFFIX);

					if ((reject_msg != NULL) &&
						(reject_msg[0] != '\0')) {

						if ((rfd = open(name_buf,
							O_RDWR|O_CREAT|O_TRUNC, 0600)) == -1) {
							sprintf(log_buffer,
								"open of reject file %s failed: errno %d",
								name_buf, errno);
							log_event(PBSEVENT_JOB,
								PBS_EVENTCLASS_JOB,
								LOG_INFO, jobp->ji_qs.ji_jobid,
								log_buffer);
						} else {
#ifdef WIN32
							secure_file(name_buf, "Administrators",
								READS_MASK|WRITES_MASK|STANDARD_RIGHTS_REQUIRED);
							setmode(rfd, O_BINARY);
#endif
							len = strlen(reject_msg)+1;
							/* write also trailing null char */
							if (write(rfd, reject_msg, len) != len) {
								sprintf(log_buffer,
									"write to file %s incomplete: errno %d", name_buf, errno);
								log_event(PBSEVENT_JOB,
									PBS_EVENTCLASS_JOB,
									LOG_INFO, jobp->ji_qs.ji_jobid,
									log_buffer);
							}
							close(rfd);
						}
					}

					if (err == PBSE_HOOKERROR)
						exit(SEND_JOB_HOOKERR);
					if (err == PBSE_HOOK_REJECT)
						exit(SEND_JOB_HOOK_REJECT);
					if (err == PBSE_HOOK_REJECT_RERUNJOB)
						exit(SEND_JOB_HOOK_REJECT_RERUNJOB);
					if (err == PBSE_HOOK_REJECT_DELETEJOB)
						exit(SEND_JOB_HOOK_REJECT_DELETEJOB);
				}
				else {
					(void)sprintf(log_buffer,
						"send of job to %s failed error = %d",
						destin, pbs_errno);
					log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB,
						LOG_INFO, jobp->ji_qs.ji_jobid,
						log_buffer);
					continue;
				}
			}

			if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
				if (PBSD_jscript(con, script_name, rpp, NULL) != 0)
					continue;
			}

			if (credlen > 0) {
				int	ret;

				ret = PBSD_jcred(con,
					jobp->ji_extended.ji_ext.ji_credtype,
					credbuf, credlen, rpp, NULL);
				if ((ret == 0) || (i == (RETRY - 1)))
					free(credbuf);	/* free credbuf if cred info is sent successfully OR */
				/* at the end of all retry attempts */
				if (ret != 0)
					continue;
			}

			if ((move_type == MOVE_TYPE_Exec) &&
				(jobp->ji_qs.ji_svrflags & JOB_SVFLG_HASRUN) &&
				(hostaddr !=  pbs_server_addr)) {
				/* send files created on prior run */
				if ((move_job_file(con, jobp, StdOut, rpp, NULL) != 0) ||
					(move_job_file(con, jobp, StdErr, rpp, NULL) != 0) ||
					(move_job_file(con, jobp, Chkpt, rpp, NULL) != 0))
					continue;
			}

			jobp->ji_qs.ji_substate = JOB_SUBSTATE_TRNOUTCM;
		}

		if (PBSD_rdytocmt(con, job_id, rpp, NULL) != 0)
			continue;

		if (PBSD_commit(con, job_id, rpp, NULL) != 0) {
			/* delete the temp script file */
			unlink(script_name);
			exit(SEND_JOB_FATAL);
		}
		svr_disconnect(con);

		/* delete the temp script file */
		unlink(script_name);

		exit(SEND_JOB_OK);	/* This child process is all done */
	}
	if (con >= 0)
		svr_disconnect(con);
	/*
	 * If connection is actively refused by the execution node(or mother superior) OR
	 * the execution node(or mother superior) is rejecting request with error
	 * PBSE_BADHOST(failing to authorize server host), the node should be marked down.
	 */
	if ((move_type == MOVE_TYPE_Exec) && (pbs_errno == ECONNREFUSED  || pbs_errno == PBSE_BADHOST)) {
		i = SEND_JOB_NODEDW;
	} else if (should_retry_route(pbs_errno) == -1) {
		i = SEND_JOB_FATAL;
	} else {
		i = SEND_JOB_RETRY;
	}
	(void)sprintf(log_buffer, "send_job failed with error %d", pbs_errno);
	log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB, LOG_NOTICE,
		jobp->ji_qs.ji_jobid, log_buffer);

	/* delete the temp script file */
	unlink(script_name);

	exit(i);
	return -1;		/* NOT REACHED */

#endif /* !WIN32 */
}
Esempio n. 2
0
/**
 * @brief
 * 		This is the Windows counterpart of the monitoring
 *		code.
 * @par
 *		This function does the following:
 *		a) Creates/opens a file $PBS_HOME/datastore/pbs_dblock.
 *		b) If mode is "check", attempts to lock the file. If locking
 *		succeeds, unlocks the file and returns success.
 *		c) If mode is "monitor", launches itself with a "monitorchild"
 *		parameter, which calls function "win_db_monitor_child".
 *		d) It launches a child process using win_popen() and reads its stdout.
 *		e) If the child was able to successfully lock the file, it prints "0"
 *	   to its stdout. Otherwise it prints the reason for why it could
 *	   not acquire the lockfile.
 *
 * @param[in]	mode	-	"check"	: to just check if lockfile can be locked
 *		     				"monitor"	:  to launch a monitoring process that holds
 *				 			onto the file lock
 *
 * @retval	1	: Function failed to acquire lock
 * @retval	0	: Function succeded in the requested operation
 *
 * @par MT-safe:	Yes
 */
int
win_db_monitor(char *mode)
{
	int rc;
	BOOL fSuccess = FALSE;
	HANDLE hFile;
	char lockfile[MAXPATHLEN + 1];
	char cmd_line[2*MAXPATHLEN + 1];
	char result[RES_BUF_SIZE];
	int is_lock_local = 0;
	pio_handles pio;
	proc_ctrl proc_info;
	HANDLE hOut, hErr;

	result[0] = '\0';
	snprintf(lockfile, MAXPATHLEN, "%s\\datastore\\pbs_dblock", pbs_conf.pbs_home_path);

	/*
	 * If mode is check, just attempt to lock the file.
	 * Return success if able to lock, else return failure.
	 */
	if (strcmp(mode, "check") == 0) {
		hFile = acquire_lock(lockfile, result, sizeof(result), &is_lock_local);
		if (hFile == INVALID_HANDLE_VALUE) {
			if (is_lock_local)
				return 0; /* Since lock is already held by this host, return success */
			fprintf(stderr, "Failed to acquire lock on %s. %s\n", lockfile, result);
			return 1;
		}

		lock_out(hFile, F_UNLCK);
		CloseHandle(hFile);
		unlink(lockfile);
		return 0;
	}

	/* monitor part */
	proc_info.flags = CREATE_DEFAULT_ERROR_MODE | CREATE_NO_WINDOW;
	proc_info.bInheritHandle = TRUE;
	proc_info.bnowait = TRUE;
	proc_info.need_ptree_termination = FALSE;
	proc_info.buse_cmd = FALSE;

	sprintf(cmd_line, "\"%s\" monitorchild", pbs_ds_monitor_exe);

	/* set the current processes stdout/stderr not be inherited */
	hOut = GetStdHandle(STD_OUTPUT_HANDLE);
	SetHandleInformation(hOut, HANDLE_FLAG_INHERIT, 0);

	hErr = GetStdHandle(STD_ERROR_HANDLE);
	SetHandleInformation(hErr, HANDLE_FLAG_INHERIT, 0);

	/* start child process to lock db lockfile and monitor db process */
	if (win_popen(cmd_line, "r", &pio, &proc_info) == 0) {
		win_pclose(&pio);
		fprintf(stderr, "Unable to create process, errno = %d\n", errno);
		return 1;
	}

	/* wait and read the info from child whether it was able to acquire lock */
	rc = win_pread(&pio, result, sizeof(result) - 1);

	win_pclose2(&pio); /* close handles but keep process running */

	if (rc > 0) {
		if (result[0] == '0') { /* indicates success */
			return 0;
		}
		result[rc - 1] = '\0';
	}

	/* failure */
	fprintf(stderr, "Failed to acquire lock on %s. %s\n", lockfile, result);
	return 1;
}
Esempio n. 3
0
/*
 * @brief
 *      PBS_authenticate - call pbs_iff(1) to authenticate use to the PBS server.
 *
 * @note
 * This function now accepts a argument sock_port and invoke pbs_iff
 * passing this port as a command line argument (both on unix and windows)
 * This change is done because getsockname() fails sometimes on Windows.
 *
 * Also, this would create an environment variable PBS_IFF_CLIENT_ADDR set to
 * the client's connecting address, which is made known to the pbs_iff process.
 *
 * If unable to authenticate, an attempt is made to run the old method
 * 'pbs_iff -i <pbs_client_addr>' also.
 *
 *
 * @param[in]  psock           Socket descriptor used by PBS client to connect PBS server.
 * @param[in]  server_name     Connecting PBS server host name.
 * @param[in]  server_port     Connecting PBS server port number.
 * @param[in]  paddr           Connecting PBS client sockaddr.
 *
 * @return int
 * @retval  0 on success.
 * @retval -1 on failure.
 */
static int
PBSD_authenticate(int psock, char * server_name, int server_port,
	struct sockaddr_in *paddr)
{
	char   cmd[2][PBS_MAXSERVERNAME + 80];
	int    cred_type;
	int    i, k;
	char*  pbs_client_addr = NULL;
	u_short psock_port = 0;
	int	rc;
#ifdef WIN32
	struct	pio_handles	pio;
#else
	FILE	*piff;
#endif
	if (paddr == NULL) {
		return (-1);
	}
	pbs_client_addr = inet_ntoa(paddr->sin_addr);
	if (pbs_client_addr == NULL) {
		return (-1);
	}
	psock_port = paddr->sin_port;

	/* for compatibility with 12.0 pbs_iff */
	(void)snprintf(cmd[1], sizeof(cmd[1])-1, "%s -i %s %s %u %d %u",
		pbs_conf.iff_path, pbs_client_addr,
		server_name, server_port, psock, psock_port);
#ifdef WIN32

	(void)snprintf(cmd[0], sizeof(cmd[0])-1, "%s %s %u %d %u", pbs_conf.iff_path,
		server_name, server_port, psock, psock_port);
	for (k=0; k < 2; k++) {
		rc = 0;
		SetEnvironmentVariable(PBS_IFF_CLIENT_ADDR, pbs_client_addr);
		if (!win_popen(cmd[k], "r", &pio, NULL)) {
			printf("failed to execute %s\n", cmd[k]);
			SetEnvironmentVariable(PBS_IFF_CLIENT_ADDR, NULL);
			rc = -1;
			break;
		}
		i=win_pread(&pio, (char *)&cred_type, (int)sizeof(int));
		win_pclose(&pio);
		SetEnvironmentVariable(PBS_IFF_CLIENT_ADDR, NULL);

		if ((i != sizeof(int)) ||
			(cred_type != PBS_credentialtype_none)) {
			rc = -1;
		} else {
			break;
		}
	}

#else	/* UNIX code here */
	/* Use pbs_iff to authenticate me */
	(void)snprintf(cmd[0], sizeof(cmd[0])-1, "%s=%s %s %s %u %d %u", PBS_IFF_CLIENT_ADDR,
		pbs_client_addr, pbs_conf.iff_path,
		server_name, server_port, psock, psock_port);

	for (k=0; k < 2; k++) {
		rc = 0;

		piff = (FILE *)popen(cmd[k], "r");
		if (piff == NULL) {
			rc = -1;
			break;
		}

		while ((i = read(fileno(piff), &cred_type, sizeof(int))) == -1) {
			if (errno != EINTR)
				break;
		}

		(void)pclose(piff);
		if ((i != sizeof(int)) ||
			(cred_type != PBS_credentialtype_none)) {
			rc = -1;
		} else {
			break;
		}
	}
#endif	/* end of UNIX code */

	return rc;
}