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;
}
extern int slurm_persist_msg_unpack(slurm_persist_conn_t *persist_conn,
persist_msg_t *resp_msg, Buf buffer)
{
int rc;
xassert(persist_conn);
xassert(resp_msg);
if (persist_conn->flags & PERSIST_FLAG_DBD) {
rc = unpack_slurmdbd_msg((slurmdbd_msg_t *)resp_msg,
persist_conn->version,
buffer);
} else {
slurm_msg_t msg;
slurm_msg_t_init(&msg);
msg.protocol_version = persist_conn->version;
safe_unpack16(&msg.msg_type, buffer);
rc = unpack_msg(&msg, buffer);
resp_msg->msg_type = msg.msg_type;
resp_msg->data = msg.data;
}
/* Here we transfer the auth_cred to the persist_conn just in case in the
* future we need to use it in some way to verify things for messages
* that don't have on that will follow on the connection.
*/
if (resp_msg->msg_type == REQUEST_PERSIST_INIT) {
slurm_msg_t *msg = resp_msg->data;
if (persist_conn->auth_cred)
g_slurm_auth_destroy(persist_conn->auth_cred);
persist_conn->auth_cred = msg->auth_cred;
msg->auth_cred = NULL;
}
return rc;
unpack_error:
return SLURM_ERROR;
}
/* Send an RPC to the SlurmDBD and wait for an arbitrary reply message.
* The RPC will not be queued if an error occurs.
* The "resp" message must be freed by the caller.
* Returns SLURM_SUCCESS or an error code */
extern int send_recv_slurmdbd_msg(uint16_t rpc_version,
slurmdbd_msg_t *req,
slurmdbd_msg_t *resp)
{
int rc = SLURM_SUCCESS;
Buf buffer;
xassert(req);
xassert(resp);
/* To make sure we can get this to send instead of the agent
sending stuff that can happen anytime we set halt_agent and
then after we get into the mutex we unset.
*/
halt_agent = 1;
slurm_mutex_lock(&slurmdbd_lock);
halt_agent = 0;
if (!slurmdbd_conn || (slurmdbd_conn->fd < 0)) {
/* Either slurm_open_slurmdbd_conn() was not executed or
* the connection to Slurm DBD has been closed */
if (req->msg_type == DBD_GET_CONFIG)
_open_slurmdbd_conn(0);
else
_open_slurmdbd_conn(1);
if (!slurmdbd_conn || (slurmdbd_conn->fd < 0)) {
rc = SLURM_ERROR;
goto end_it;
}
}
if (!(buffer = pack_slurmdbd_msg(req, rpc_version))) {
rc = SLURM_ERROR;
goto end_it;
}
rc = slurm_persist_send_msg(slurmdbd_conn, buffer);
free_buf(buffer);
if (rc != SLURM_SUCCESS) {
error("slurmdbd: Sending message type %s: %d: %m",
rpc_num2string(req->msg_type), rc);
goto end_it;
}
buffer = slurm_persist_recv_msg(slurmdbd_conn);
if (buffer == NULL) {
error("slurmdbd: Getting response to message type %u",
req->msg_type);
rc = SLURM_ERROR;
goto end_it;
}
rc = unpack_slurmdbd_msg(resp, rpc_version, buffer);
/* check for the rc of the start job message */
if (rc == SLURM_SUCCESS && resp->msg_type == DBD_ID_RC)
rc = ((dbd_id_rc_msg_t *)resp->data)->return_code;
free_buf(buffer);
end_it:
slurm_cond_signal(&slurmdbd_cond);
slurm_mutex_unlock(&slurmdbd_lock);
return rc;
}
static void _load_dbd_state(void)
{
char *dbd_fname;
Buf buffer;
int fd, recovered = 0;
uint16_t rpc_version = 0;
dbd_fname = slurm_get_state_save_location();
xstrcat(dbd_fname, "/dbd.messages");
fd = open(dbd_fname, O_RDONLY);
if (fd < 0) {
/* don't print an error message if there is no file */
if (errno == ENOENT)
debug4("slurmdbd: There is no state save file to "
"open by name %s", dbd_fname);
else
error("slurmdbd: Opening state save file %s: %m",
dbd_fname);
} else {
char *ver_str = NULL;
uint32_t ver_str_len;
buffer = _load_dbd_rec(fd);
if (buffer == NULL)
goto end_it;
/* This is set to the end of the buffer for send so we
need to set it back to 0 */
set_buf_offset(buffer, 0);
safe_unpackstr_xmalloc(&ver_str, &ver_str_len, buffer);
debug3("Version string in dbd_state header is %s", ver_str);
unpack_error:
free_buf(buffer);
buffer = NULL;
if (ver_str) {
/* get the version after VER */
rpc_version = slurm_atoul(ver_str + 3);
xfree(ver_str);
}
while (1) {
/* If the buffer was not the VER%d string it
was an actual message so we don't want to
skip it.
*/
if (!buffer)
buffer = _load_dbd_rec(fd);
if (buffer == NULL)
break;
if (rpc_version != SLURM_PROTOCOL_VERSION) {
/* unpack and repack with new
* PROTOCOL_VERSION just so we keep
* things up to date.
*/
slurmdbd_msg_t msg;
int rc;
set_buf_offset(buffer, 0);
rc = unpack_slurmdbd_msg(
&msg, rpc_version, buffer);
free_buf(buffer);
if (rc == SLURM_SUCCESS)
buffer = pack_slurmdbd_msg(
&msg, SLURM_PROTOCOL_VERSION);
else
buffer = NULL;
}
if (!buffer) {
error("no buffer given");
continue;
}
if (!list_enqueue(agent_list, buffer))
fatal("slurmdbd: list_enqueue, no memory");
recovered++;
buffer = NULL;
}
end_it:
verbose("slurmdbd: recovered %d pending RPCs", recovered);
(void) close(fd);
}
xfree(dbd_fname);
}
