/*
* Receive the first message identifying initiator
*/
static void
_direct_conn_establish(pmixp_conn_t *conn, void *_hdr, void *msg)
{
pmixp_io_engine_t *eng = pmixp_conn_get_eng(conn);
pmixp_base_hdr_t *hdr = (pmixp_base_hdr_t *)_hdr;
pmixp_dconn_t *dconn = NULL;
pmixp_conn_t *new_conn;
eio_obj_t *obj;
int fd;
fd = pmixp_io_detach(eng);
dconn = pmixp_dconn_accept(hdr->nodeid, fd);
if (!dconn) {
/* connection was refused because we already
* have established connection
* It seems that some sort of race condition occured
*/
char *nodename = pmixp_info_job_host(hdr->nodeid);
close(fd);
PMIXP_ERROR("Failed to accept direct connection from %s",
nodename);
xfree(nodename);
return;
}
new_conn = pmixp_conn_new_persist(PMIXP_PROTO_DIRECT,
pmixp_dconn_engine(dconn),
_direct_new_msg_conn,
_direct_return_connection, dconn);
pmixp_dconn_unlock(dconn);
obj = eio_obj_create(fd, &direct_peer_ops, (void *)new_conn);
eio_new_obj(pmixp_info_io(), obj);
/* wakeup this connection to get processed */
eio_signal_wakeup(pmixp_info_io());
}
void pmixp_server_direct_conn(int fd)
{
eio_obj_t *obj;
pmixp_conn_t *conn;
PMIXP_DEBUG("Request from fd = %d", fd);
/* Set nonblocking */
fd_set_nonblocking(fd);
fd_set_close_on_exec(fd);
pmixp_fd_set_nodelay(fd);
conn = pmixp_conn_new_temp(PMIXP_PROTO_DIRECT, fd,
_direct_conn_establish);
/* try to process right here */
pmixp_conn_progress_rcv(conn);
if (!pmixp_conn_is_alive(conn)) {
/* success, don't need this connection anymore */
pmixp_conn_return(conn);
return;
}
/* If it is a blocking operation: create AIO object to
* handle it */
obj = eio_obj_create(fd, &direct_peer_ops, (void *)conn);
eio_new_obj(pmixp_info_io(), obj);
/* wakeup this connection to get processed */
eio_signal_wakeup(pmixp_info_io());
}
/*
* TODO: we need to keep track of the "me"
* structures created here, because we need to
* free them in "pmixp_stepd_finalize"
*/
void pmixp_server_slurm_conn(int fd)
{
eio_obj_t *obj;
pmixp_conn_t *conn = NULL;
PMIXP_DEBUG("Request from fd = %d", fd);
pmixp_debug_hang(0);
/* Set nonblocking */
fd_set_nonblocking(fd);
fd_set_close_on_exec(fd);
conn = pmixp_conn_new_temp(PMIXP_PROTO_SLURM, fd, _slurm_new_msg);
/* try to process right here */
pmixp_conn_progress_rcv(conn);
if (!pmixp_conn_is_alive(conn)) {
/* success, don't need this connection anymore */
pmixp_conn_return(conn);
return;
}
/* If it is a blocking operation: create AIO object to
* handle it */
obj = eio_obj_create(fd, &slurm_peer_ops, (void *)conn);
eio_new_obj(pmixp_info_io(), obj);
}
/*
* TODO: we need to keep track of the "me"
* structures created here, because we need to
* free them in "pmixp_stepd_finalize"
*/
void pmix_server_new_conn(int fd)
{
eio_obj_t *obj;
PMIXP_DEBUG("Request from fd = %d", fd);
/* Set nonblocking */
fd_set_nonblocking(fd);
fd_set_close_on_exec(fd);
pmixp_io_engine_t *me = xmalloc(sizeof(pmixp_io_engine_t));
pmix_io_init(me, fd, srv_rcvd_header);
/* We use slurm_forward_data to send message to stepd's
* SLURM will put user ID there. We need to skip it.
*/
pmix_io_rcvd_padding(me, sizeof(uint32_t));
if( 2 == _process_message(me) ){
/* connection was fully processed here */
xfree(me);
return;
}
/* If it is a blocking operation: create AIO object to
* handle it */
obj = eio_obj_create(fd, &peer_ops, (void *)me);
eio_new_obj(pmixp_info_io(), obj);
}
static int _process_extended_hdr(pmixp_base_hdr_t *hdr, Buf buf)
{
char nhdr[PMIXP_BASE_HDR_MAX];
bool send_init = false;
size_t dsize = 0, hsize = 0;
pmixp_dconn_t *dconn;
_direct_proto_message_t *init_msg = NULL;
int rc = SLURM_SUCCESS;
char *ep_data = NULL;
uint32_t ep_len = 0;
dconn = pmixp_dconn_lock(hdr->nodeid);
if (!dconn) {
/* Should not happen */
xassert( dconn );
abort();
}
/* Retrieve endpoint information */
_base_hdr_unpack_ext(buf, &ep_data, &ep_len);
/* Check if init message is required to establish
* the connection
*/
if (!pmixp_dconn_require_connect(dconn, &send_init)) {
goto unlock;
}
if (send_init) {
Buf buf_init = pmixp_server_buf_new();
pmixp_base_hdr_t bhdr;
init_msg = xmalloc(sizeof(*init_msg));
PMIXP_BASE_HDR_SETUP(bhdr, PMIXP_MSG_INIT_DIRECT, 0, buf_init);
bhdr.ext_flag = 1;
hsize = _direct_hdr_pack(&bhdr, nhdr);
init_msg->sent_cb = pmixp_server_sent_buf_cb;
init_msg->cbdata = buf_init;
init_msg->hdr = bhdr;
init_msg->buffer = _buf_finalize(buf_init, nhdr, hsize,
&dsize);
init_msg->buf_ptr = buf_init;
}
rc = pmixp_dconn_connect(dconn, ep_data, ep_len, init_msg);
if (rc) {
PMIXP_ERROR("Unable to connect to %d", dconn->nodeid);
if (init_msg) {
/* need to release `init_msg` here */
free_buf(init_msg->buf_ptr);
xfree(init_msg);
}
goto unlock;
}
switch (pmixp_dconn_progress_type(dconn)) {
case PMIXP_DCONN_PROGRESS_SW:{
/* this direct connection has fd that needs to be
* polled to progress, use connection interface for that
*/
pmixp_io_engine_t *eng = pmixp_dconn_engine(dconn);
pmixp_conn_t *conn;
conn = pmixp_conn_new_persist(PMIXP_PROTO_DIRECT, eng,
_direct_new_msg_conn,
_direct_return_connection,
dconn);
if (conn) {
eio_obj_t *obj;
obj = eio_obj_create(pmixp_io_fd(eng),
&direct_peer_ops,
(void *)conn);
eio_new_obj(pmixp_info_io(), obj);
eio_signal_wakeup(pmixp_info_io());
} else {
/* TODO: handle this error */
rc = SLURM_ERROR;
goto unlock;
}
break;
}
case PMIXP_DCONN_PROGRESS_HW: {
break;
}
default:
/* Should not happen */
xassert(0 && pmixp_dconn_progress_type(dconn));
/* TODO: handle this error */
}
unlock:
pmixp_dconn_unlock(dconn);
return rc;
}