/*
* Receives message data.
* @param msg the message to be received into
* @param peer the peer to receive from
* @retval true if the whole message was received
* @retval false if the whole message was not received
*/
bool mca_oob_tcp_msg_recv_handler(mca_oob_tcp_msg_t* msg, struct mca_oob_tcp_peer_t * peer)
{
/* has entire header been received */
if(msg->msg_rwptr == msg->msg_rwiov) {
if(mca_oob_tcp_msg_recv(msg, peer) == false)
return false;
/* allocate a buffer for the receive */
MCA_OOB_TCP_HDR_NTOH(&msg->msg_hdr);
if(msg->msg_hdr.msg_size > 0) {
msg->msg_rwbuf = malloc(msg->msg_hdr.msg_size);
if(NULL == msg->msg_rwbuf) {
opal_output(0, "%s-%s mca_oob_tcp_msg_recv_handler: malloc(%d) failed\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->peer_name)),
msg->msg_hdr.msg_size);
mca_oob_tcp_peer_close(peer);
return false;
}
msg->msg_rwiov[1].iov_base = (ompi_iov_base_ptr_t)msg->msg_rwbuf;
msg->msg_rwiov[1].iov_len = msg->msg_hdr.msg_size;
msg->msg_rwnum = 1;
} else {
msg->msg_rwiov[1].iov_base = NULL;
msg->msg_rwiov[1].iov_len = 0;
msg->msg_rwnum = 0;
}
if (mca_oob_tcp_component.tcp_debug >= OOB_TCP_DEBUG_INFO) {
opal_output(0, "%s-%s (origin: %s) mca_oob_tcp_msg_recv_handler: size %lu\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->peer_name)),
ORTE_NAME_PRINT(&(msg->msg_hdr.msg_origin)),
(unsigned long)(msg->msg_hdr.msg_size) );
}
}
/* do the right thing based on the message type */
switch(msg->msg_hdr.msg_type) {
case MCA_OOB_TCP_IDENT:
/* done - there is nothing else to receive */
return true;
case MCA_OOB_TCP_PING:
/* done - there is nothing else to receive */
return true;
case MCA_OOB_TCP_DATA:
/* finish receiving message */
return mca_oob_tcp_msg_recv(msg, peer);
default:
return true;
}
}
/*
* Receive the peers globally unique process identification from a newly
* connected socket and verify the expected response. If so, move the
* socket to a connected state.
*/
static int mca_oob_tcp_peer_recv_connect_ack(mca_oob_tcp_peer_t* peer)
{
mca_oob_tcp_hdr_t hdr;
if((mca_oob_tcp_peer_recv_blocking(peer, &hdr, sizeof(hdr))) != sizeof(hdr)) {
/* If the peer state is still CONNECT_ACK, that indicates that
the error was a reset from the remote host because the
connection was not able to be fully established. In that
case, Clean up the connection and give it another go. */
if (peer->peer_state == MCA_OOB_TCP_CONNECT_ACK) {
struct timeval tv = { 1,0 };
if (mca_oob_tcp_component.tcp_debug >= OOB_TCP_DEBUG_CONNECT) {
opal_output(0,
"[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_peer_recv_connect_ack "
"connect failed during receive. Restarting (%s).",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(&(peer->peer_name)),
strerror(opal_socket_errno));
}
opal_event_del(&peer->peer_recv_event);
mca_oob_tcp_peer_shutdown(peer);
opal_evtimer_add(&peer->peer_timer_event, &tv);
return ORTE_SUCCESS;
} else {
mca_oob_tcp_peer_close(peer);
return ORTE_ERR_UNREACH;
}
}
MCA_OOB_TCP_HDR_NTOH(&hdr);
if(hdr.msg_type != MCA_OOB_TCP_CONNECT) {
opal_output(0, "mca_oob_tcp_peer_recv_connect_ack: invalid header type: %d\n",
hdr.msg_type);
mca_oob_tcp_peer_close(peer);
return ORTE_ERR_UNREACH;
}
/* compare the peers name to the expected value */
if(memcmp(&peer->peer_name, &hdr.msg_src, sizeof(orte_process_name_t)) != 0) {
opal_output(0, "[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_peer_recv_connect_ack: "
"received unexpected process identifier [%d,%d,%d]\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(&(peer->peer_name)),
ORTE_NAME_ARGS(&(hdr.msg_src)));
mca_oob_tcp_peer_close(peer);
return ORTE_ERR_UNREACH;
}
/* if we have an invalid name or do not have one assigned at all -
* use the name returned by the peer. This needs to be a LITERAL
* comparison - we do NOT want wildcard values to return EQUAL
*/
if(orte_process_info.my_name == NULL) {
orte_ns.create_process_name(&orte_process_info.my_name,
hdr.msg_dst.cellid, hdr.msg_dst.jobid, hdr.msg_dst.vpid);
} else if (orte_ns.compare_fields(ORTE_NS_CMP_ALL, orte_process_info.my_name, ORTE_NAME_INVALID) == ORTE_EQUAL) {
*orte_process_info.my_name = hdr.msg_dst;
}
/* connected */
mca_oob_tcp_peer_connected(peer);
if(mca_oob_tcp_component.tcp_debug >= OOB_TCP_DEBUG_CONNECT) {
mca_oob_tcp_peer_dump(peer, "connected");
}
return ORTE_SUCCESS;
}
void mca_oob_tcp_component_hop_unknown(int fd, short args, void *cbdata)
{
mca_oob_tcp_msg_error_t *mop = (mca_oob_tcp_msg_error_t*)cbdata;
uint64_t ui64;
orte_rml_send_t *snd;
orte_oob_base_peer_t *bpr;
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s tcp:unknown hop called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->hop));
if (orte_finalizing || orte_abnormal_term_ordered) {
/* just ignore the problem */
OBJ_RELEASE(mop);
return;
}
/* mark that this component cannot reach this hop */
memcpy(&ui64, (char*)&(mop->hop), sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) ||
NULL == bpr) {
/* the overall OOB has no knowledge of this hop. Only
* way this could happen is if the peer contacted us
* via this component, and it wasn't entered into the
* OOB framework hash table. We have no way of knowing
* what to do next, so just output an error message and
* abort */
opal_output(0, "%s ERROR: message to %s requires routing and the OOB has no knowledge of the reqd hop %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->snd->hdr.dst),
ORTE_NAME_PRINT(&mop->hop));
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_COMM_FAILED);
OBJ_RELEASE(mop);
return;
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
/* mark that this component cannot reach this destination either */
memcpy(&ui64, (char*)&(mop->snd->hdr.dst), sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&orte_oob_base.peers,
ui64, (void**)&bpr) ||
NULL == bpr) {
opal_output(0, "%s ERROR: message to %s requires routing and the OOB has no knowledge of this process",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&mop->snd->hdr.dst));
ORTE_ACTIVATE_PROC_STATE(&mop->hop, ORTE_PROC_STATE_COMM_FAILED);
OBJ_RELEASE(mop);
return;
}
opal_bitmap_clear_bit(&bpr->addressable, mca_oob_tcp_component.super.idx);
/* post the message to the OOB so it can see
* if another component can transfer it
*/
MCA_OOB_TCP_HDR_NTOH(&mop->snd->hdr);
snd = OBJ_NEW(orte_rml_send_t);
snd->dst = mop->snd->hdr.dst;
snd->origin = mop->snd->hdr.origin;
snd->tag = mop->snd->hdr.tag;
snd->data = mop->snd->data;
snd->count = mop->snd->hdr.nbytes;
snd->cbfunc.iov = NULL;
snd->cbdata = NULL;
/* activate the OOB send state */
ORTE_OOB_SEND(snd);
/* protect the data */
mop->snd->data = NULL;
OBJ_RELEASE(mop);
}
void mca_oob_tcp_recv_handler(int sd, short flags, void *cbdata)
{
mca_oob_tcp_peer_t* peer = (mca_oob_tcp_peer_t*)cbdata;
int rc;
orte_process_name_t hop;
mca_oob_tcp_peer_t *relay;
uint64_t ui64;
if (orte_abnormal_term_ordered) {
return;
}
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name));
switch (peer->state) {
case MCA_OOB_TCP_CONNECT_ACK:
if (ORTE_SUCCESS == (rc = mca_oob_tcp_peer_recv_connect_ack(peer, peer->sd, NULL))) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler starting send/recv events",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* we connected! Start the send/recv events */
if (!peer->recv_ev_active) {
opal_event_add(&peer->recv_event, 0);
peer->recv_ev_active = true;
}
if (peer->timer_ev_active) {
opal_event_del(&peer->timer_event);
peer->timer_ev_active = false;
}
/* if there is a message waiting to be sent, queue it */
if (NULL == peer->send_msg) {
peer->send_msg = (mca_oob_tcp_send_t*)opal_list_remove_first(&peer->send_queue);
}
if (NULL != peer->send_msg && !peer->send_ev_active) {
opal_event_add(&peer->send_event, 0);
peer->send_ev_active = true;
}
/* update our state */
peer->state = MCA_OOB_TCP_CONNECTED;
} else {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s UNABLE TO COMPLETE CONNECT ACK WITH %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name));
opal_event_del(&peer->recv_event);
ORTE_FORCED_TERMINATE(1);
return;
}
break;
case MCA_OOB_TCP_CONNECTED:
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler CONNECTED",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* allocate a new message and setup for recv */
if (NULL == peer->recv_msg) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler allocate new recv msg",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
peer->recv_msg = OBJ_NEW(mca_oob_tcp_recv_t);
if (NULL == peer->recv_msg) {
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: unable to allocate recv message\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)));
return;
}
/* start by reading the header */
peer->recv_msg->rdptr = (char*)&peer->recv_msg->hdr;
peer->recv_msg->rdbytes = sizeof(mca_oob_tcp_hdr_t);
}
/* if the header hasn't been completely read, read it */
if (!peer->recv_msg->hdr_recvd) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler read hdr",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
if (ORTE_SUCCESS == (rc = read_bytes(peer))) {
OPAL_TIMING_EVENT((&tm,"Header received from %s",
ORTE_NAME_PRINT(&peer->name)));
/* completed reading the header */
peer->recv_msg->hdr_recvd = true;
/* convert the header */
MCA_OOB_TCP_HDR_NTOH(&peer->recv_msg->hdr);
/* if this is a zero-byte message, then we are done */
if (0 == peer->recv_msg->hdr.nbytes) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s RECVD ZERO-BYTE MESSAGE FROM %s for tag %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name), peer->recv_msg->hdr.tag);
peer->recv_msg->data = NULL; // make sure
} else {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler allocate data region of size %lu",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (unsigned long)peer->recv_msg->hdr.nbytes);
/* allocate the data region */
peer->recv_msg->data = (char*)malloc(peer->recv_msg->hdr.nbytes);
/* point to it */
peer->recv_msg->rdptr = peer->recv_msg->data;
peer->recv_msg->rdbytes = peer->recv_msg->hdr.nbytes;
}
/* fall thru and attempt to read the data */
} else if (ORTE_ERR_RESOURCE_BUSY == rc ||
ORTE_ERR_WOULD_BLOCK == rc) {
/* exit this event and let the event lib progress */
return;
} else {
/* close the connection */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler error reading bytes - closing connection",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
mca_oob_tcp_peer_close(peer);
return;
}
}
if (peer->recv_msg->hdr_recvd) {
/* continue to read the data block - we start from
* wherever we left off, which could be at the
* beginning or somewhere in the message
*/
if (ORTE_SUCCESS == (rc = read_bytes(peer))) {
OPAL_TIMING_EVENT((&tm,"Msg received from %s",
ORTE_NAME_PRINT(&peer->name)));
/* we recvd all of the message */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s RECVD COMPLETE MESSAGE FROM %s (ORIGIN %s) OF %d BYTES FOR DEST %s TAG %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name),
ORTE_NAME_PRINT(&peer->recv_msg->hdr.origin),
(int)peer->recv_msg->hdr.nbytes,
ORTE_NAME_PRINT(&peer->recv_msg->hdr.dst),
peer->recv_msg->hdr.tag);
/* am I the intended recipient (header was already converted back to host order)? */
if (peer->recv_msg->hdr.dst.jobid == ORTE_PROC_MY_NAME->jobid &&
peer->recv_msg->hdr.dst.vpid == ORTE_PROC_MY_NAME->vpid) {
/* yes - post it to the RML for delivery */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s DELIVERING TO RML",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
ORTE_RML_POST_MESSAGE(&peer->recv_msg->hdr.origin, peer->recv_msg->hdr.tag,
peer->recv_msg->data,
peer->recv_msg->hdr.nbytes);
OBJ_RELEASE(peer->recv_msg);
} else {
/* no - find the next hop in the route */
hop = orte_routed.get_route(&peer->recv_msg->hdr.dst);
if (hop.jobid == ORTE_JOBID_INVALID ||
hop.vpid == ORTE_VPID_INVALID) {
/* no hop known - post the error to the component
* and let the OOB see if there is another way
* to get there from here
*/
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s NO ROUTE TO %s FROM HERE",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name));
/* let the component know about the problem */
ORTE_ACTIVATE_TCP_MSG_ERROR(NULL, peer->recv_msg, &hop, mca_oob_tcp_component_no_route);
/* cleanup */
OBJ_RELEASE(peer->recv_msg);
return;
} else {
/* does we know how to reach the next hop? */
memcpy(&ui64, (char*)&hop, sizeof(uint64_t));
if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&mca_oob_tcp_module.peers, ui64, (void**)&relay)) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s ADDRESS OF NEXT HOP %s TO %s IS UNKNOWN",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&hop),
ORTE_NAME_PRINT(&peer->recv_msg->hdr.dst));
/* let the component know about the problem */
ORTE_ACTIVATE_TCP_MSG_ERROR(NULL, peer->recv_msg, &hop, mca_oob_tcp_component_hop_unknown);
/* cleanup */
OBJ_RELEASE(peer->recv_msg);
return;
}
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s ROUTING TO %s FROM HERE",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&relay->name));
/* if this came from a different job family, then ensure
* we know how to return
*/
if (ORTE_JOB_FAMILY(peer->recv_msg->hdr.origin.jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
orte_routed.update_route(&(peer->recv_msg->hdr.origin), &peer->name);
}
/* post the message for retransmission */
MCA_OOB_TCP_QUEUE_RELAY(peer->recv_msg, relay);
OBJ_RELEASE(peer->recv_msg);
}
}
peer->recv_msg = NULL;
return;
} else if (ORTE_ERR_RESOURCE_BUSY == rc ||
ORTE_ERR_WOULD_BLOCK == rc) {
/* exit this event and let the event lib progress */
return;
} else {
// report the error
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: unable to recv message",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)));
/* turn off the recv event */
opal_event_del(&peer->recv_event);
ORTE_FORCED_TERMINATE(1);
return;
}
}
break;
default:
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: invalid socket state(%d)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)),
peer->state);
// mca_oob_tcp_peer_close(peer);
break;
}
}
void mca_oob_tcp_recv_handler(int sd, short flags, void *cbdata)
{
mca_oob_tcp_peer_t* peer = (mca_oob_tcp_peer_t*)cbdata;
int rc;
orte_rml_send_t *snd;
if (orte_abnormal_term_ordered) {
return;
}
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler called for peer %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name));
switch (peer->state) {
case MCA_OOB_TCP_CONNECT_ACK:
if (ORTE_SUCCESS == (rc = mca_oob_tcp_peer_recv_connect_ack(peer, peer->sd, NULL))) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler starting send/recv events",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* we connected! Start the send/recv events */
if (!peer->recv_ev_active) {
opal_event_add(&peer->recv_event, 0);
peer->recv_ev_active = true;
}
if (peer->timer_ev_active) {
opal_event_del(&peer->timer_event);
peer->timer_ev_active = false;
}
/* if there is a message waiting to be sent, queue it */
if (NULL == peer->send_msg) {
peer->send_msg = (mca_oob_tcp_send_t*)opal_list_remove_first(&peer->send_queue);
}
if (NULL != peer->send_msg && !peer->send_ev_active) {
opal_event_add(&peer->send_event, 0);
peer->send_ev_active = true;
}
/* update our state */
peer->state = MCA_OOB_TCP_CONNECTED;
} else {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s UNABLE TO COMPLETE CONNECT ACK WITH %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name));
opal_event_del(&peer->recv_event);
ORTE_FORCED_TERMINATE(1);
return;
}
break;
case MCA_OOB_TCP_CONNECTED:
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler CONNECTED",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* allocate a new message and setup for recv */
if (NULL == peer->recv_msg) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler allocate new recv msg",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
peer->recv_msg = OBJ_NEW(mca_oob_tcp_recv_t);
if (NULL == peer->recv_msg) {
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: unable to allocate recv message\n",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)));
return;
}
/* start by reading the header */
peer->recv_msg->rdptr = (char*)&peer->recv_msg->hdr;
peer->recv_msg->rdbytes = sizeof(mca_oob_tcp_hdr_t);
}
/* if the header hasn't been completely read, read it */
if (!peer->recv_msg->hdr_recvd) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler read hdr",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
if (ORTE_SUCCESS == (rc = read_bytes(peer))) {
/* completed reading the header */
peer->recv_msg->hdr_recvd = true;
/* convert the header */
MCA_OOB_TCP_HDR_NTOH(&peer->recv_msg->hdr);
/* if this is a zero-byte message, then we are done */
if (0 == peer->recv_msg->hdr.nbytes) {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s RECVD ZERO-BYTE MESSAGE FROM %s for tag %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name), peer->recv_msg->hdr.tag);
peer->recv_msg->data = NULL; // make sure
} else {
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler allocate data region of size %lu",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (unsigned long)peer->recv_msg->hdr.nbytes);
/* allocate the data region */
peer->recv_msg->data = (char*)malloc(peer->recv_msg->hdr.nbytes);
/* point to it */
peer->recv_msg->rdptr = peer->recv_msg->data;
peer->recv_msg->rdbytes = peer->recv_msg->hdr.nbytes;
}
/* fall thru and attempt to read the data */
} else if (ORTE_ERR_RESOURCE_BUSY == rc ||
ORTE_ERR_WOULD_BLOCK == rc) {
/* exit this event and let the event lib progress */
return;
} else {
/* close the connection */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s:tcp:recv:handler error reading bytes - closing connection",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
mca_oob_tcp_peer_close(peer);
return;
}
}
if (peer->recv_msg->hdr_recvd) {
/* continue to read the data block - we start from
* wherever we left off, which could be at the
* beginning or somewhere in the message
*/
if (ORTE_SUCCESS == (rc = read_bytes(peer))) {
/* we recvd all of the message */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s RECVD COMPLETE MESSAGE FROM %s (ORIGIN %s) OF %d BYTES FOR DEST %s TAG %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->name),
ORTE_NAME_PRINT(&peer->recv_msg->hdr.origin),
(int)peer->recv_msg->hdr.nbytes,
ORTE_NAME_PRINT(&peer->recv_msg->hdr.dst),
peer->recv_msg->hdr.tag);
/* am I the intended recipient (header was already converted back to host order)? */
if (peer->recv_msg->hdr.dst.jobid == ORTE_PROC_MY_NAME->jobid &&
peer->recv_msg->hdr.dst.vpid == ORTE_PROC_MY_NAME->vpid) {
/* yes - post it to the RML for delivery */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s DELIVERING TO RML",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
ORTE_RML_POST_MESSAGE(&peer->recv_msg->hdr.origin, peer->recv_msg->hdr.tag,
peer->recv_msg->data,
peer->recv_msg->hdr.nbytes);
OBJ_RELEASE(peer->recv_msg);
} else {
/* promote this to the OOB as some other transport might
* be the next best hop */
opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
"%s TCP PROMOTING ROUTED MESSAGE FOR %s TO OOB",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&peer->recv_msg->hdr.dst));
snd = OBJ_NEW(orte_rml_send_t);
snd->dst = peer->recv_msg->hdr.dst;
snd->origin = peer->recv_msg->hdr.origin;
snd->tag = peer->recv_msg->hdr.tag;
snd->data = peer->recv_msg->data;
snd->count = peer->recv_msg->hdr.nbytes;
snd->cbfunc.iov = NULL;
snd->cbdata = NULL;
/* activate the OOB send state */
ORTE_OOB_SEND(snd);
/* protect the data */
peer->recv_msg->data = NULL;
/* cleanup */
OBJ_RELEASE(peer->recv_msg);
}
peer->recv_msg = NULL;
return;
} else if (ORTE_ERR_RESOURCE_BUSY == rc ||
ORTE_ERR_WOULD_BLOCK == rc) {
/* exit this event and let the event lib progress */
return;
} else {
// report the error
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: unable to recv message",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)));
/* turn off the recv event */
opal_event_del(&peer->recv_event);
ORTE_FORCED_TERMINATE(1);
return;
}
}
break;
default:
opal_output(0, "%s-%s mca_oob_tcp_peer_recv_handler: invalid socket state(%d)",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&(peer->name)),
peer->state);
// mca_oob_tcp_peer_close(peer);
break;
}
}
int mca_oob_tcp_ping(
const orte_process_name_t* name,
const char* uri,
const struct timeval *timeout)
{
int sd, flags, rc;
struct sockaddr_in inaddr;
fd_set fdset;
mca_oob_tcp_hdr_t hdr;
struct timeval tv;
struct iovec iov;
#ifndef __WINDOWS__
struct opal_event sigpipe_handler;
#endif
/* parse uri string */
if(ORTE_SUCCESS != (rc = mca_oob_tcp_parse_uri(uri, &inaddr))) {
opal_output(0,
"[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_ping: invalid uri: %s\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(name),
uri);
return rc;
}
/* create socket */
sd = socket(AF_INET, SOCK_STREAM, 0);
if (sd < 0) {
opal_output(0,
"[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_ping: socket() failed: %s (%d)\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(name),
strerror(opal_socket_errno),
opal_socket_errno);
return ORTE_ERR_UNREACH;
}
/* setup the socket as non-blocking */
if((flags = fcntl(sd, F_GETFL, 0)) < 0) {
opal_output(0, "[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_ping: fcntl(F_GETFL) failed: %s (%d)\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(name),
strerror(opal_socket_errno),
opal_socket_errno);
} else {
flags |= O_NONBLOCK;
if(fcntl(sd, F_SETFL, flags) < 0) {
opal_output(0, "[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_ping: fcntl(F_SETFL) failed: %s (%d)\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(name),
strerror(opal_socket_errno),
opal_socket_errno);
}
}
/* start the connect - will likely fail with EINPROGRESS */
FD_ZERO(&fdset);
if(connect(sd, (struct sockaddr*)&inaddr, sizeof(inaddr)) < 0) {
/* connect failed? */
if(opal_socket_errno != EINPROGRESS && opal_socket_errno != EWOULDBLOCK) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
/* select with timeout to wait for connect to complete */
FD_SET(sd, &fdset);
tv = *timeout;
rc = select(sd+1, NULL, &fdset, NULL, &tv);
if(rc <= 0) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
}
/* set socket back to blocking */
flags &= ~O_NONBLOCK;
if(fcntl(sd, F_SETFL, flags) < 0) {
opal_output(0, "[%lu,%lu,%lu]-[%lu,%lu,%lu] mca_oob_tcp_ping: fcntl(F_SETFL) failed: %s (%d)\n",
ORTE_NAME_ARGS(orte_process_info.my_name),
ORTE_NAME_ARGS(name),
strerror(opal_socket_errno),
opal_socket_errno);
}
/* send a probe message */
memset(&hdr, 0, sizeof(hdr));
if(orte_process_info.my_name != NULL) {
hdr.msg_src = *orte_process_info.my_name;
} else {
hdr.msg_src = *ORTE_NAME_INVALID;
}
hdr.msg_dst = *name;
hdr.msg_type = MCA_OOB_TCP_PROBE;
MCA_OOB_TCP_HDR_HTON(&hdr);
#ifndef __WINDOWS__
/* Ignore SIGPIPE in the write -- determine success or failure in
the ping by looking at the return code from write() */
opal_signal_set(&sigpipe_handler, SIGPIPE,
noop, &sigpipe_handler);
opal_signal_add(&sigpipe_handler, NULL);
#endif
/* Do the write and see what happens. Use the writev version just to
* make Windows happy as there the write function is limitted to
* file operations.
*/
iov.iov_base = (IOVBASE_TYPE*)&hdr;
iov.iov_len = sizeof(hdr);
rc = writev(sd, &iov, 1 );
#ifndef __WINDOWS__
/* Now de-register the handler */
opal_signal_del(&sigpipe_handler);
#endif
if (rc != sizeof(hdr)) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
/* select with timeout to wait for response */
FD_SET(sd, &fdset);
tv = *timeout;
rc = select(sd+1, &fdset, NULL, NULL, &tv);
if(rc <= 0) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
if((rc = read(sd, &hdr, sizeof(hdr))) != sizeof(hdr)) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
MCA_OOB_TCP_HDR_NTOH(&hdr);
if(hdr.msg_type != MCA_OOB_TCP_PROBE) {
CLOSE_THE_SOCKET(sd);
return ORTE_ERR_UNREACH;
}
CLOSE_THE_SOCKET(sd);
return ORTE_SUCCESS;
}