C++ slurm_seterrno函数代码示例

/*
 * slurm_create_reservation - create a new reservation, only usable by user root
 * IN resv_msg - description of reservation
 * RET name of reservation on success (caller must free the memory),
 *	otherwise return NULL and set errno to indicate the error
 */
char *
slurm_create_reservation (resv_desc_msg_t * resv_msg)
{
	int rc;
	char *resv_name = NULL;
	slurm_msg_t req_msg;
	slurm_msg_t resp_msg;
	reservation_name_msg_t *resp;

	slurm_msg_t_init(&req_msg);
	slurm_msg_t_init(&resp_msg);

	req_msg.msg_type = REQUEST_CREATE_RESERVATION;
	req_msg.data     = resv_msg;

	rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg);
	switch (resp_msg.msg_type) {
	case RESPONSE_CREATE_RESERVATION:
		resp = (reservation_name_msg_t *) resp_msg.data;
		resv_name = strdup(resp->name);
		break;
	case RESPONSE_SLURM_RC:
		rc = ((return_code_msg_t *) resp_msg.data)->return_code;
		if (rc)
			slurm_seterrno(rc);
		break;
	default:
		slurm_seterrno(SLURM_UNEXPECTED_MSG_ERROR);
	}
	slurm_free_msg_data(resp_msg.msg_type, resp_msg.data);
	return resv_name;
}

/*
 * _suspend_op2 - perform a suspend/resume operation for some job.
 * IN op         - operation to perform
 * IN job_id_str - job on which to perform operation in string format or NULL
 * OUT resp      - slurm error codes by job array task ID
 * RET 0 or a slurm error code
 * NOTE: Supply either job_id NO_VAL or job_id_str as NULL, not both
 */
static int _suspend_op2(uint16_t op, char *job_id_str,
			job_array_resp_msg_t **resp)
{
	int rc = SLURM_SUCCESS;
	suspend_msg_t sus_req;
	slurm_msg_t req_msg, resp_msg;

	slurm_msg_t_init(&req_msg);
	slurm_msg_t_init(&resp_msg);
	sus_req.op         = op;
	sus_req.job_id     = NO_VAL;
	sus_req.job_id_str = job_id_str;
	req_msg.msg_type   = REQUEST_SUSPEND;
	req_msg.data       = &sus_req;

	rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg);
	switch (resp_msg.msg_type) {
	case RESPONSE_JOB_ARRAY_ERRORS:
		*resp = (job_array_resp_msg_t *) resp_msg.data;
		break;
	case RESPONSE_SLURM_RC:
		rc = ((return_code_msg_t *) resp_msg.data)->return_code;
		if (rc)
			slurm_seterrno(rc);
		break;
	default:
		slurm_seterrno(SLURM_UNEXPECTED_MSG_ERROR);
	}

	return rc;
}

/*
 * slurm_update_job2 - issue RPC to a job's configuration per request,
 *	only usable by user root or (for some parameters) the job's owner
 * IN job_msg - description of job updates
 * OUT resp - per task response to the request,
 *	      free using slurm_free_job_array_resp()
 * RET 0 on success, otherwise return -1 and set errno to indicate the error
 */
extern int
slurm_update_job2 (job_desc_msg_t * job_msg, job_array_resp_msg_t **resp)
{
	int rc = SLURM_SUCCESS;
	slurm_msg_t req_msg, resp_msg;

	slurm_msg_t_init(&req_msg);
	slurm_msg_t_init(&resp_msg);
	req_msg.msg_type	= REQUEST_UPDATE_JOB;
	req_msg.data		= job_msg;

	rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg);
	switch (resp_msg.msg_type) {
	case RESPONSE_JOB_ARRAY_ERRORS:
		*resp = (job_array_resp_msg_t *) resp_msg.data;
		break;
	case RESPONSE_SLURM_RC:
		rc = ((return_code_msg_t *) resp_msg.data)->return_code;
		if (rc)
			slurm_seterrno(rc);
		break;
	default:
		slurm_seterrno(SLURM_UNEXPECTED_MSG_ERROR);
	}

	return rc;
}

/*
 * slurm_requeue2 - re-queue a batch job, if already running
 *	then terminate it first
 * IN job_id_str - job on which to perform operation in string format or NULL
 * IN state      - state in which to place the job
 * RET 0 or a slurm error code
 */
extern int slurm_requeue2(char *job_id_str, uint32_t state,
			  job_array_resp_msg_t **resp)
{
	int rc = SLURM_SUCCESS;
	requeue_msg_t requeue_req;
	slurm_msg_t req_msg, resp_msg;

	slurm_msg_t_init(&req_msg);
	slurm_msg_t_init(&resp_msg);
	requeue_req.job_id	= NO_VAL;
	requeue_req.job_id_str	= job_id_str;
	requeue_req.state	= state;
	req_msg.msg_type	= REQUEST_JOB_REQUEUE;
	req_msg.data		= &requeue_req;

	rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg);
	switch (resp_msg.msg_type) {
	case RESPONSE_JOB_ARRAY_ERRORS:
		*resp = (job_array_resp_msg_t *) resp_msg.data;
		break;
	case RESPONSE_SLURM_RC:
		rc = ((return_code_msg_t *) resp_msg.data)->return_code;
		if (rc)
			slurm_seterrno(rc);
		break;
	default:
		slurm_seterrno(SLURM_UNEXPECTED_MSG_ERROR);
	}

	return rc;
}

/* returns 0 if invalid gid, otherwise returns 1.  Set gid with
 * correct gid if root launched job.  Also set user_name
 * if not already set. */
static int
_valid_uid_gid(uid_t uid, gid_t *gid, char **user_name)
{
	struct passwd *pwd;
	struct group *grp;
	int i;

#ifdef HAVE_NATIVE_CRAY
	/* already verified */
	if (*user_name)
		return 1;
#endif

	pwd = getpwuid(uid);
	if (!pwd) {
		error("uid %ld not found on system", (long) uid);
		slurm_seterrno(ESLURMD_UID_NOT_FOUND);
		return 0;
	}

	if (!*user_name)
		*user_name = xstrdup(pwd->pw_name);

	if (pwd->pw_gid == *gid)
		return 1;

	grp = getgrgid(*gid);
	if (!grp) {
		error("gid %ld not found on system", (long)(*gid));
		slurm_seterrno(ESLURMD_GID_NOT_FOUND);
		return 0;
	}

	/* Allow user root to use any valid gid */
	if (pwd->pw_uid == 0) {
		pwd->pw_gid = *gid;
		return 1;
	}
	for (i = 0; grp->gr_mem[i]; i++) {
		if (!strcmp(pwd->pw_name, grp->gr_mem[i])) {
			pwd->pw_gid = *gid;
		       	return 1;
	       	}
	}

	/* root user may have launched this job for this user, but
	 * root did not explicitly set the gid. This would set the
	 * gid to 0. In this case we should set the appropriate
	 * default gid for the user (from the passwd struct).
	 */
	if (*gid == 0) {
		*gid = pwd->pw_gid;
		return 1;
	}
	error("uid %ld is not a member of gid %ld",
		(long)pwd->pw_uid, (long)(*gid));
	slurm_seterrno(ESLURMD_GID_NOT_FOUND);

	return 0;
}

slurm_fd_t _slurm_open_stream(slurm_addr_t *addr, bool retry)
{
	int retry_cnt;
	slurm_fd_t fd;
	uint16_t port;
	char ip[32];

	if ( (addr->sin_family == 0) || (addr->sin_port  == 0) ) {
		error("Error connecting, bad data: family = %u, port = %u",
			addr->sin_family, addr->sin_port);
		return SLURM_SOCKET_ERROR;
	}

	for (retry_cnt=0; ; retry_cnt++) {
		int rc;
		if ((fd =_slurm_create_socket(SLURM_STREAM)) < 0) {
			error("Error creating slurm stream socket: %m");
			slurm_seterrno(errno);
			return SLURM_SOCKET_ERROR;
		}

		if (retry_cnt) {
			if (retry_cnt == 1) {
				debug3("Error connecting, "
				       "picking new stream port");
			}
			_sock_bind_wild(fd);
		}

		rc = _slurm_connect(fd, (struct sockaddr const *)addr,
				    sizeof(*addr));
		if (rc >= 0)		    /* success */
			break;
		if (((errno != ECONNREFUSED) && (errno != ETIMEDOUT)) ||
		    (!retry) || (retry_cnt >= PORT_RETRIES)) {
			slurm_seterrno(errno);
			goto error;
		}

		if ((_slurm_close_stream(fd) < 0) && (errno == EINTR))
			_slurm_close_stream(fd);	/* try again */
	}

	return fd;

    error:
	slurm_get_ip_str(addr, &port, ip, sizeof(ip));
	debug2("Error connecting slurm stream socket at %s:%d: %m",
	       ip, ntohs(port));
	if ((_slurm_close_stream(fd) < 0) && (errno == EINTR))
		_slurm_close_stream(fd);	/* try again */
	return SLURM_SOCKET_ERROR;
}

/*
 * slurm_update_job2 - issue RPC to a job's configuration per request,
 *	only usable by user root or (for some parameters) the job's owner
 * IN job_msg - description of job updates
 * OUT resp - per task response to the request,
 *	      free using slurm_free_job_array_resp()
 * RET SLURM_SUCCESS on success, otherwise return SLURM_ERROR with errno set
 */
extern int
slurm_update_job2 (job_desc_msg_t * job_msg, job_array_resp_msg_t **resp)
{
	int rc = SLURM_SUCCESS;
	slurm_msg_t req_msg, resp_msg;
	slurmdb_cluster_rec_t *save_working_cluster_rec = working_cluster_rec;

	slurm_msg_t_init(&req_msg);
	req_msg.msg_type	= REQUEST_UPDATE_JOB;
	req_msg.data		= job_msg;

tryagain:
	slurm_msg_t_init(&resp_msg);

	rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg,
					    working_cluster_rec);
	switch (resp_msg.msg_type) {
	case RESPONSE_SLURM_REROUTE_MSG:
	{
		reroute_msg_t *rr_msg = (reroute_msg_t *)resp_msg.data;

		/* Don't expect mutliple hops but in the case it does
		 * happen, free the previous rr cluster_rec. */
		if (working_cluster_rec &&
		    working_cluster_rec != save_working_cluster_rec)
			slurmdb_destroy_cluster_rec(
						working_cluster_rec);

		working_cluster_rec = rr_msg->working_cluster_rec;
		slurmdb_setup_cluster_rec(working_cluster_rec);
		rr_msg->working_cluster_rec = NULL;
		goto tryagain;
	}
	case RESPONSE_JOB_ARRAY_ERRORS:
		*resp = (job_array_resp_msg_t *) resp_msg.data;
		break;
	case RESPONSE_SLURM_RC:
		rc = ((return_code_msg_t *) resp_msg.data)->return_code;
		if (rc)
			slurm_seterrno(rc);
		break;
	default:
		slurm_seterrno(SLURM_UNEXPECTED_MSG_ERROR);
	}

	if (working_cluster_rec != save_working_cluster_rec) {
		slurmdb_destroy_cluster_rec(working_cluster_rec);
		working_cluster_rec = save_working_cluster_rec;
	}

	return rc;
}

/* slurm_load_assoc_mgr_info()
 *
 * Load requested controller assoc_mgr state.
 *
 */
extern int
slurm_load_assoc_mgr_info(assoc_mgr_info_request_msg_t *req,
			  assoc_mgr_info_msg_t **resp)
{
	int cc;
	slurm_msg_t msg_request;
	slurm_msg_t msg_reply;

	slurm_msg_t_init(&msg_request);
	slurm_msg_t_init(&msg_reply);

	msg_request.msg_type = REQUEST_ASSOC_MGR_INFO;
	msg_request.data = req;

	cc = slurm_send_recv_controller_msg(&msg_request, &msg_reply);
	if (cc < 0)
		return SLURM_ERROR;

	switch (msg_reply.msg_type) {
		case RESPONSE_ASSOC_MGR_INFO:
			*resp = msg_reply.data;
			break;
		case RESPONSE_SLURM_RC:
			cc = ((return_code_msg_t *)msg_reply.data)->return_code;
			slurm_free_return_code_msg(msg_reply.data);
			if (cc) /* slurm_seterrno_ret() is a macro ... sigh */
				slurm_seterrno(cc);
			return -1;
		default:
			slurm_seterrno_ret(SLURM_UNEXPECTED_MSG_ERROR);
		break;
	}

	return SLURM_PROTOCOL_SUCCESS;
}

extern int
proctrack_p_wait(uint64_t cont_id)
{
	pid_t pgid = (pid_t)cont_id;
	int delay = 1;

	if (cont_id == 0 || cont_id == 1) {
		slurm_seterrno(EINVAL);
		return SLURM_ERROR;
	}

	/* Spin until the process group is gone. */
	while (killpg(pgid, 0) == 0) {
		proctrack_p_signal(cont_id, SIGKILL);
		sleep(delay);
		if (delay < 120) {
			delay *= 2;
		} else {
			error("%s: Unable to destroy container %"PRIu64" "
			      "in pgid plugin, giving up after %d sec",
			      __func__, cont_id, delay);
			break;
		}
	}

	return SLURM_SUCCESS;
}

extern int checkpoint_tasks (uint32_t job_id, uint32_t step_id,
			     time_t begin_time, char *image_dir,
			     uint16_t wait, char *nodelist)
{
	int rc = SLURM_SUCCESS, temp_rc;
	checkpoint_tasks_msg_t ckpt_req;
	slurm_msg_t req_msg;
	List ret_list;
        ret_data_info_t *ret_data_info = NULL;

	slurm_msg_t_init(&req_msg);
	ckpt_req.job_id		= job_id;
	ckpt_req.job_step_id 	= step_id;
	ckpt_req.timestamp	= begin_time;
	ckpt_req.image_dir	= image_dir;
	req_msg.msg_type	= REQUEST_CHECKPOINT_TASKS;
	req_msg.data		= &ckpt_req;

	if ((ret_list = slurm_send_recv_msgs(nodelist, &req_msg, (wait*1000),
					     false))) {
		while ((ret_data_info = list_pop(ret_list))) {
                        temp_rc = slurm_get_return_code(ret_data_info->type,
                                                        ret_data_info->data);
                        if (temp_rc)
                                rc = temp_rc;
                }
	} else {
                error("slurm_checkpoint_tasks: no list was returned");
                rc = SLURM_ERROR;
	}
	slurm_seterrno(rc);
	return rc;
}

/*
 *  Handle a return code message type.
 *    Sets errno to return code and returns it
 */
static int
_handle_rc_msg(slurm_msg_t *msg)
{
	int rc = ((return_code_msg_t *) msg->data)->return_code;
	slurm_free_return_code_msg(msg->data);
	slurm_seterrno(rc);
	return rc;
}

extern int acct_gather_check_acct_freq_task(uint64_t job_mem_lim,
					    char *acctg_freq)
{
	int task_freq;
	static uint32_t acct_freq_task = NO_VAL;

	if (acct_freq_task == NO_VAL) {
		char *acct_freq = slurm_get_jobacct_gather_freq();
		int i = acct_gather_parse_freq(PROFILE_TASK, acct_freq);
		xfree(acct_freq);

		/* If the value is -1 lets set the freq to something
		   really high so we don't check this again.
		*/
		if (i == -1)
			acct_freq_task = (uint16_t)NO_VAL;
		else
			acct_freq_task = i;
	}

	if (!job_mem_lim || !acct_freq_task)
		return 0;

	task_freq = acct_gather_parse_freq(PROFILE_TASK, acctg_freq);

	if (task_freq == -1)
		return 0;

	if (task_freq == 0) {
		error("Can't turn accounting frequency off.  "
		      "We need it to monitor memory usage.");
		slurm_seterrno(ESLURMD_INVALID_ACCT_FREQ);
		return 1;
	} else if (task_freq > acct_freq_task) {
		error("Can't set frequency to %d, it is higher than %u.  "
		      "We need it to be at least at this level to "
		      "monitor memory usage.",
		      task_freq, acct_freq_task);
		slurm_seterrno(ESLURMD_INVALID_ACCT_FREQ);
		return 1;
	}

	return 0;
}

/*
 * scontrol_requeue - requeue a pending or running batch job
 * IN job_id_str - a job id
 */
extern void
scontrol_requeue(char *job_str)
{
	char *job_id_str;
	int rc, i;
	job_array_resp_msg_t *resp = NULL;

	if (!job_str[0]) {
		exit_code = 1;
		return;
	}

	if (xstrncasecmp(job_str, "jobid=", 6) == 0)
		job_str += 6;
	if (xstrncasecmp(job_str, "job=", 4) == 0)
		job_str += 4;

	if (_is_job_id(job_str)) {
		job_id_str = _next_job_id();
		while (job_id_str) {
			rc = slurm_requeue2(job_id_str, 0, &resp);
			if (rc != SLURM_SUCCESS) {
				exit_code = 1;
				if (quiet_flag != 1) {
					fprintf(stderr, "%s for job %s\n",
						slurm_strerror(slurm_get_errno()),
						job_id_str);
				}
			} else if (resp) {
				for (i = 0; i < resp->job_array_count; i++) {
					if ((resp->error_code[i] == SLURM_SUCCESS)
					    && (resp->job_array_count == 1))
						continue;
					exit_code = 1;
					if (quiet_flag == 1)
						continue;
					fprintf(stderr, "%s: %s\n",
						resp->job_array_id[i],
						slurm_strerror(resp->
							       error_code[i]));
				}
				slurm_free_job_array_resp(resp);
				resp = NULL;
			}
			job_id_str = _next_job_id();
		}
	} else {
		exit_code = 1;
		rc = ESLURM_INVALID_JOB_ID;
		slurm_seterrno(rc);
		if (quiet_flag != 1) {
			fprintf(stderr, "%s for job %s\n",
				slurm_strerror(rc), job_str);
		}
	}
}

extern int proctrack_p_signal  ( uint64_t id, int signal )
{
	pid_t pid = (pid_t) id;

	if (!id) {
		/* no container ID */
	} else if (pid == getpid() || pid == getpgid(0)) {
		error("slurm_signal_container would kill caller!");
	} else {
		return killpg(pid, signal);
	}
	slurm_seterrno(ESRCH);
	return SLURM_ERROR;
}

/*
 * scontrol_hold - perform some job hold/release operation
 * IN op - suspend/resume operation
 * IN job_id_str - a job id
 * RET 0 if no slurm error, errno otherwise. parsing error prints
 *		error message and returns 0
 */
extern int
scontrol_hold(char *op, char *job_id_str)
{
	int rc = SLURM_SUCCESS;
	char *next_str;
	job_desc_msg_t job_msg;
	uint16_t job_state;

	slurm_init_job_desc_msg (&job_msg);

	/* set current user, needed e.g., for AllowGroups checks */
	job_msg.user_id = getuid();

	if (job_id_str) {
		job_msg.job_id = (uint32_t) strtol(job_id_str, &next_str, 10);
		if ((job_msg.job_id == 0) || (next_str[0] != '\0')) {
			fprintf(stderr, "Invalid job id specified\n");
			exit_code = 1;
			return 0;
		}
	} else {
		fprintf(stderr, "Invalid job id specified\n");
		exit_code = 1;
		return 0;
	}

	job_state = scontrol_get_job_state(job_msg.job_id);
	if (job_state == (uint16_t) NO_VAL)
		return SLURM_ERROR;
	if ((job_state & JOB_STATE_BASE) != JOB_PENDING) {
		slurm_seterrno(ESLURM_JOB_NOT_PENDING);
		return ESLURM_JOB_NOT_PENDING;
	}

	if ((strncasecmp(op, "holdu", 5) == 0) ||
	    (strncasecmp(op, "uhold", 5) == 0)) {
		job_msg.priority = 0;
		job_msg.alloc_sid = ALLOC_SID_USER_HOLD;
	} else if (strncasecmp(op, "hold", 4) == 0) {
		job_msg.priority = 0;
		job_msg.alloc_sid = 0;
	} else
		job_msg.priority = INFINITE;

	if (slurm_update_job(&job_msg))
		return slurm_get_errno();

	return rc;
}