extern int slurm_persist_unpack_rc_msg(
persist_rc_msg_t **msg, Buf buffer, uint16_t protocol_version)
{
uint32_t uint32_tmp;
persist_rc_msg_t *msg_ptr = xmalloc(sizeof(persist_rc_msg_t));
*msg = msg_ptr;
if (protocol_version >= SLURM_MIN_PROTOCOL_VERSION) {
safe_unpackstr_xmalloc(&msg_ptr->comment, &uint32_tmp, buffer);
safe_unpack16(&msg_ptr->flags, buffer);
safe_unpack32(&msg_ptr->rc, buffer);
safe_unpack16(&msg_ptr->ret_info, buffer);
} else {
error("%s: invalid protocol_version %u",
__func__, protocol_version);
goto unpack_error;
}
return SLURM_SUCCESS;
unpack_error:
slurm_persist_free_rc_msg(msg_ptr);
*msg = NULL;
return SLURM_ERROR;
}
static int _unpack_return_code(uint16_t rpc_version, Buf buffer)
{
uint16_t msg_type = -1;
persist_rc_msg_t *msg;
dbd_id_rc_msg_t *id_msg;
slurmdbd_msg_t resp;
int rc = SLURM_ERROR;
memset(&resp, 0, sizeof(slurmdbd_msg_t));
if ((rc = unpack_slurmdbd_msg(&resp, slurmdbd_conn->version, buffer))
!= SLURM_SUCCESS) {
error("%s: unpack message error", __func__);
return rc;
}
switch (resp.msg_type) {
case DBD_ID_RC:
id_msg = resp.data;
rc = id_msg->return_code;
slurmdbd_free_id_rc_msg(id_msg);
if (rc != SLURM_SUCCESS)
error("slurmdbd: DBD_ID_RC is %d", rc);
break;
case PERSIST_RC:
msg = resp.data;
rc = msg->rc;
if (rc != SLURM_SUCCESS) {
if (msg->ret_info == DBD_REGISTER_CTLD &&
slurm_get_accounting_storage_enforce()) {
error("slurmdbd: PERSIST_RC is %d from "
"%s(%u): %s",
rc,
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info,
msg->comment);
fatal("You need to add this cluster "
"to accounting if you want to "
"enforce associations, or no "
"jobs will ever run.");
} else
debug("slurmdbd: PERSIST_RC is %d from "
"%s(%u): %s",
rc,
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info,
msg->comment);
}
slurm_persist_free_rc_msg(msg);
break;
default:
error("slurmdbd: bad message type %d != PERSIST_RC", msg_type);
}
return rc;
}
/* Send an RPC to the SlurmDBD and wait for the return code reply.
* The RPC will not be queued if an error occurs.
* Returns SLURM_SUCCESS or an error code */
extern int send_slurmdbd_recv_rc_msg(uint16_t rpc_version,
slurmdbd_msg_t *req,
int *resp_code)
{
int rc;
slurmdbd_msg_t resp;
xassert(req);
xassert(resp_code);
memset(&resp, 0, sizeof(slurmdbd_msg_t));
rc = send_recv_slurmdbd_msg(rpc_version, req, &resp);
if (rc != SLURM_SUCCESS) {
; /* error message already sent */
} else if (resp.msg_type != PERSIST_RC) {
error("slurmdbd: response is not type PERSIST_RC: %s(%u)",
slurmdbd_msg_type_2_str(resp.msg_type, 1),
resp.msg_type);
rc = SLURM_ERROR;
} else { /* resp.msg_type == PERSIST_RC */
persist_rc_msg_t *msg = resp.data;
*resp_code = msg->rc;
if (msg->rc != SLURM_SUCCESS &&
msg->rc != ACCOUNTING_FIRST_REG &&
msg->rc != ACCOUNTING_TRES_CHANGE_DB &&
msg->rc != ACCOUNTING_NODES_CHANGE_DB) {
char *comment = msg->comment;
if (!comment)
comment = slurm_strerror(msg->rc);
if (msg->ret_info == DBD_REGISTER_CTLD &&
slurm_get_accounting_storage_enforce()) {
error("slurmdbd: Issue with call "
"%s(%u): %u(%s)",
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info, msg->rc,
comment);
fatal("You need to add this cluster "
"to accounting if you want to "
"enforce associations, or no "
"jobs will ever run.");
} else
debug("slurmdbd: Issue with call "
"%s(%u): %u(%s)",
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info, msg->rc,
comment);
}
slurm_persist_free_rc_msg(msg);
}
return rc;
}
extern int slurm_persist_unpack_rc_msg(
persist_rc_msg_t **msg, Buf buffer, uint16_t protocol_version)
{
uint32_t uint32_tmp;
persist_rc_msg_t *msg_ptr = xmalloc(sizeof(persist_rc_msg_t));
*msg = msg_ptr;
safe_unpackstr_xmalloc(&msg_ptr->comment, &uint32_tmp, buffer);
safe_unpack32(&msg_ptr->rc, buffer);
safe_unpack16(&msg_ptr->ret_info, buffer);
return SLURM_SUCCESS;
unpack_error:
slurm_persist_free_rc_msg(msg_ptr);
*msg = NULL;
return SLURM_ERROR;
}
/* Open a persistent socket connection
* IN/OUT - persistent connection needing rem_host and rem_port filled in.
* Returned completely filled in.
* Returns SLURM_SUCCESS on success or SLURM_ERROR on failure */
extern int slurm_persist_conn_open(slurm_persist_conn_t *persist_conn)
{
int rc = SLURM_ERROR;
slurm_msg_t req_msg;
persist_init_req_msg_t req;
persist_rc_msg_t *resp = NULL;
if (slurm_persist_conn_open_without_init(persist_conn) != SLURM_SUCCESS)
return rc;
slurm_msg_t_init(&req_msg);
/* Always send the lowest protocol since we don't know what version the
* other side is running yet.
*/
req_msg.protocol_version = persist_conn->version;
req_msg.msg_type = REQUEST_PERSIST_INIT;
req_msg.flags |= SLURM_GLOBAL_AUTH_KEY;
if (persist_conn->flags & PERSIST_FLAG_DBD)
req_msg.flags |= SLURMDBD_CONNECTION;
memset(&req, 0, sizeof(persist_init_req_msg_t));
req.cluster_name = persist_conn->cluster_name;
req.persist_type = persist_conn->persist_type;
req.port = persist_conn->my_port;
req.version = SLURM_PROTOCOL_VERSION;
req_msg.data = &req;
if (slurm_send_node_msg(persist_conn->fd, &req_msg) < 0) {
error("%s: failed to send persistent connection init message to %s:%d",
__func__, persist_conn->rem_host, persist_conn->rem_port);
_close_fd(&persist_conn->fd);
} else {
Buf buffer = slurm_persist_recv_msg(persist_conn);
persist_msg_t msg;
slurm_persist_conn_t persist_conn_tmp;
if (!buffer) {
if (_comm_fail_log(persist_conn)) {
error("%s: No response to persist_init",
__func__);
}
_close_fd(&persist_conn->fd);
goto end_it;
}
memset(&msg, 0, sizeof(persist_msg_t));
memcpy(&persist_conn_tmp, persist_conn,
sizeof(slurm_persist_conn_t));
/* The first unpack is done the same way for dbd or normal
* communication . */
persist_conn_tmp.flags &= (~PERSIST_FLAG_DBD);
rc = slurm_persist_msg_unpack(&persist_conn_tmp, &msg, buffer);
free_buf(buffer);
resp = (persist_rc_msg_t *)msg.data;
if (resp && (rc == SLURM_SUCCESS)) {
rc = resp->rc;
persist_conn->version = resp->ret_info;
persist_conn->flags |= resp->flags;
}
if (rc != SLURM_SUCCESS) {
if (resp) {
error("%s: Something happened with the receiving/processing of the persistent connection init message to %s:%d: %s",
__func__, persist_conn->rem_host,
persist_conn->rem_port, resp->comment);
} else {
error("%s: Failed to unpack persistent connection init resp message from %s:%d",
__func__,
persist_conn->rem_host,
persist_conn->rem_port);
}
_close_fd(&persist_conn->fd);
}
}
end_it:
slurm_persist_free_rc_msg(resp);
return rc;
}
static int _handle_mult_rc_ret(void)
{
Buf buffer;
uint16_t msg_type;
persist_rc_msg_t *msg = NULL;
dbd_list_msg_t *list_msg = NULL;
int rc = SLURM_ERROR;
Buf out_buf = NULL;
buffer = slurm_persist_recv_msg(slurmdbd_conn);
if (buffer == NULL)
return rc;
safe_unpack16(&msg_type, buffer);
switch (msg_type) {
case DBD_GOT_MULT_MSG:
if (slurmdbd_unpack_list_msg(
&list_msg, slurmdbd_conn->version,
DBD_GOT_MULT_MSG, buffer)
!= SLURM_SUCCESS) {
error("slurmdbd: unpack message error");
break;
}
slurm_mutex_lock(&agent_lock);
if (agent_list) {
ListIterator itr =
list_iterator_create(list_msg->my_list);
while ((out_buf = list_next(itr))) {
Buf b;
if ((rc = _unpack_return_code(
slurmdbd_conn->version, out_buf))
!= SLURM_SUCCESS)
break;
if ((b = list_dequeue(agent_list))) {
free_buf(b);
} else {
error("slurmdbd: DBD_GOT_MULT_MSG "
"unpack message error");
}
}
list_iterator_destroy(itr);
}
slurm_mutex_unlock(&agent_lock);
slurmdbd_free_list_msg(list_msg);
break;
case PERSIST_RC:
if (slurm_persist_unpack_rc_msg(
&msg, buffer, slurmdbd_conn->version)
== SLURM_SUCCESS) {
rc = msg->rc;
if (rc != SLURM_SUCCESS) {
if (msg->ret_info == DBD_REGISTER_CTLD &&
slurm_get_accounting_storage_enforce()) {
error("slurmdbd: PERSIST_RC is %d from "
"%s(%u): %s",
rc,
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info,
msg->comment);
fatal("You need to add this cluster "
"to accounting if you want to "
"enforce associations, or no "
"jobs will ever run.");
} else
debug("slurmdbd: PERSIST_RC is %d from "
"%s(%u): %s",
rc,
slurmdbd_msg_type_2_str(
msg->ret_info, 1),
msg->ret_info,
msg->comment);
}
slurm_persist_free_rc_msg(msg);
} else
error("slurmdbd: unpack message error");
break;
default:
error("slurmdbd: bad message type %d != PERSIST_RC", msg_type);
}
unpack_error:
free_buf(buffer);
return rc;
}
