static void pmix_usock_send(int sd, short args, void *cbdata)
{
pmix_ptl_queue_t *queue = (pmix_ptl_queue_t*)cbdata;
pmix_ptl_send_t *snd;
/* acquire the object */
PMIX_ACQUIRE_OBJECT(queue);
if (NULL == queue->peer || queue->peer->sd < 0 ||
NULL == queue->peer->info || NULL == queue->peer->nptr) {
/* this peer has lost connection */
PMIX_RELEASE(queue);
/* ensure we post the object before another thread
* picks it back up */
PMIX_POST_OBJECT(queue);
return;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"[%s:%d] send to %s:%u on tag %d",
__FILE__, __LINE__,
(queue->peer)->info->pname.nspace,
(queue->peer)->info->pname.rank, (queue->tag));
snd = PMIX_NEW(pmix_ptl_send_t);
snd->hdr.pindex = htonl(pmix_globals.pindex);
snd->hdr.tag = htonl(queue->tag);
snd->hdr.nbytes = htonl((queue->buf)->bytes_used);
snd->data = (queue->buf);
/* always start with the header */
snd->sdptr = (char*)&snd->hdr;
snd->sdbytes = sizeof(pmix_ptl_hdr_t);
/* if there is no message on-deck, put this one there */
if (NULL == (queue->peer)->send_msg) {
(queue->peer)->send_msg = snd;
} else {
/* add it to the queue */
pmix_list_append(&(queue->peer)->send_queue, &snd->super);
}
/* ensure the send event is active */
if (!(queue->peer)->send_ev_active) {
(queue->peer)->send_ev_active = true;
PMIX_POST_OBJECT(queue->peer);
pmix_event_add(&(queue->peer)->send_event, 0);
}
PMIX_RELEASE(queue);
PMIX_POST_OBJECT(snd);
}
static pmix_status_t connect_to_peer(struct pmix_peer_t *peer,
pmix_info_t *info, size_t ninfo)
{
struct sockaddr_un *address;
char *evar, **uri;
pmix_status_t rc;
int sd;
pmix_socklen_t len;
bool retried = false;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"[%s:%d] connect to server",
__FILE__, __LINE__);
/* if we are not a client, there is nothing we can do */
if (!PMIX_PROC_IS_CLIENT(pmix_globals.mypeer)) {
return PMIX_ERR_NOT_SUPPORTED;
}
/* if we don't have a path to the daemon rendezvous point,
* then we need to return an error */
if (NULL != (evar = getenv("PMIX_SERVER_URI2USOCK"))) {
/* this is a v2.1+ server */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v21");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else if (NULL != (evar = getenv("PMIX_SERVER_URI"))) {
/* this is a pre-v2.1 server - must use the v12 bfrops module */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v12");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else {
/* let the caller know that the server isn't available */
return PMIX_ERR_SERVER_NOT_AVAIL;
}
/* the server will be using the same bfrops as us */
pmix_client_globals.myserver->nptr->compat.bfrops = pmix_globals.mypeer->nptr->compat.bfrops;
/* mark that we are using the V1 protocol */
pmix_globals.mypeer->protocol = PMIX_PROTOCOL_V1;
uri = pmix_argv_split(evar, ':');
if (3 != pmix_argv_count(uri)) {
pmix_argv_free(uri);
PMIX_ERROR_LOG(PMIX_ERROR);
return PMIX_ERROR;
}
/* set the server nspace */
if (NULL == pmix_client_globals.myserver->info) {
pmix_client_globals.myserver->info = PMIX_NEW(pmix_rank_info_t);
}
if (NULL == pmix_client_globals.myserver->nptr) {
pmix_client_globals.myserver->nptr = PMIX_NEW(pmix_namespace_t);
}
if (NULL == pmix_client_globals.myserver->nptr->nspace) {
pmix_client_globals.myserver->nptr->nspace = strdup(uri[0]);
}
if (NULL == pmix_client_globals.myserver->info->pname.nspace) {
pmix_client_globals.myserver->info->pname.nspace = strdup(uri[0]);
}
/* set the server rank */
pmix_client_globals.myserver->info->pname.rank = strtoull(uri[1], NULL, 10);
/* setup the path to the daemon rendezvous point */
memset(&mca_ptl_usock_component.connection, 0, sizeof(struct sockaddr_storage));
address = (struct sockaddr_un*)&mca_ptl_usock_component.connection;
address->sun_family = AF_UNIX;
snprintf(address->sun_path, sizeof(address->sun_path)-1, "%s", uri[2]);
/* if the rendezvous file doesn't exist, that's an error */
if (0 != access(uri[2], R_OK)) {
pmix_argv_free(uri);
PMIX_ERROR_LOG(PMIX_ERR_NOT_FOUND);
return PMIX_ERR_NOT_FOUND;
}
pmix_argv_free(uri);
retry:
/* establish the connection */
len = sizeof(struct sockaddr_un);
if (PMIX_SUCCESS != (rc = pmix_ptl_base_connect(&mca_ptl_usock_component.connection, len, &sd))) {
PMIX_ERROR_LOG(rc);
return rc;
}
pmix_client_globals.myserver->sd = sd;
/* send our identity and any authentication credentials to the server */
if (PMIX_SUCCESS != (rc = send_connect_ack(sd))) {
CLOSE_THE_SOCKET(sd);
return rc;
}
/* do whatever handshake is required */
if (PMIX_SUCCESS != (rc = recv_connect_ack(sd))) {
CLOSE_THE_SOCKET(sd);
if (PMIX_ERR_TEMP_UNAVAILABLE == rc) {
/* give it two tries */
if (!retried) {
retried = true;
goto retry;
}
}
return rc;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"sock_peer_try_connect: Connection across to server succeeded");
/* mark the connection as made */
pmix_globals.connected = true;
pmix_ptl_base_set_nonblocking(sd);
/* setup recv event */
pmix_event_assign(&pmix_client_globals.myserver->recv_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_READ | EV_PERSIST,
pmix_usock_recv_handler, pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver->recv_event, 0);
pmix_client_globals.myserver->recv_ev_active = true;
PMIX_POST_OBJECT(pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver->recv_event, 0);
/* setup send event */
pmix_event_assign(&pmix_client_globals.myserver->send_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_WRITE|EV_PERSIST,
pmix_usock_send_handler, pmix_client_globals.myserver);
pmix_client_globals.myserver->send_ev_active = false;
return PMIX_SUCCESS;
}
void pmix_usock_send_recv(int fd, short args, void *cbdata)
{
pmix_ptl_sr_t *ms = (pmix_ptl_sr_t*)cbdata;
pmix_ptl_posted_recv_t *req;
pmix_ptl_send_t *snd;
uint32_t tag;
/* acquire the object */
PMIX_ACQUIRE_OBJECT(ms);
if (ms->peer->sd < 0) {
/* this peer's socket has been closed */
PMIX_RELEASE(ms);
/* ensure we post the object before another thread
* picks it back up */
PMIX_POST_OBJECT(NULL);
return;
}
/* take the next tag in the sequence */
pmix_ptl_globals.current_tag++;
if (UINT32_MAX == pmix_ptl_globals.current_tag ) {
pmix_ptl_globals.current_tag = PMIX_PTL_TAG_DYNAMIC;
}
tag = pmix_ptl_globals.current_tag;
if (NULL != ms->cbfunc) {
/* if a callback msg is expected, setup a recv for it */
req = PMIX_NEW(pmix_ptl_posted_recv_t);
req->tag = tag;
req->cbfunc = ms->cbfunc;
req->cbdata = ms->cbdata;
pmix_output_verbose(5, pmix_ptl_base_framework.framework_output,
"posting recv on tag %d", req->tag);
/* add it to the list of recvs - we cannot have unexpected messages
* in this subsystem as the server never sends us something that
* we didn't previously request */
pmix_list_prepend(&pmix_ptl_globals.posted_recvs, &req->super);
}
snd = PMIX_NEW(pmix_ptl_send_t);
snd->hdr.pindex = pmix_globals.pindex;
snd->hdr.tag = tag;
snd->hdr.nbytes = ms->bfr->bytes_used;
snd->data = ms->bfr;
/* always start with the header */
snd->sdptr = (char*)&snd->hdr;
snd->sdbytes = sizeof(pmix_usock_hdr_t);
/* if there is no message on-deck, put this one there */
if (NULL == ms->peer->send_msg) {
ms->peer->send_msg = snd;
} else {
/* add it to the queue */
pmix_list_append(&ms->peer->send_queue, &snd->super);
}
/* ensure the send event is active */
if (!ms->peer->send_ev_active) {
ms->peer->send_ev_active = true;
PMIX_POST_OBJECT(snd);
pmix_event_add(&ms->peer->send_event, 0);
}
/* cleanup */
PMIX_RELEASE(ms);
PMIX_POST_OBJECT(snd);
}
static pmix_status_t connect_to_peer(struct pmix_peer_t *peer,
pmix_info_t *info, size_t ninfo)
{
char *evar, **uri, *suri;
char *filename, *nspace=NULL;
pmix_rank_t rank = PMIX_RANK_WILDCARD;
char *p, *p2;
int sd, rc;
size_t n;
char myhost[PMIX_MAXHOSTNAMELEN];
bool system_level = false;
bool system_level_only = false;
pid_t pid = 0;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp: connecting to server");
/* see if the connection info is in the info array - if
* so, then that overrides all other options */
/* if I am a client, then we need to look for the appropriate
* connection info in the environment */
if (PMIX_PROC_IS_CLIENT(pmix_globals.mypeer)) {
if (NULL != (evar = getenv("PMIX_SERVER_URI21"))) {
/* we are talking to a v2.1 server */
pmix_client_globals.myserver->proc_type = PMIX_PROC_SERVER | PMIX_PROC_V21;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"V21 SERVER DETECTED");
/* must use the v21 bfrops module */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v21");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else if (NULL != (evar = getenv("PMIX_SERVER_URI2"))) {
/* we are talking to a v2.0 server */
pmix_client_globals.myserver->proc_type = PMIX_PROC_SERVER | PMIX_PROC_V20;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"V20 SERVER DETECTED");
/* must use the v20 bfrops module */
pmix_globals.mypeer->nptr->compat.bfrops = pmix_bfrops_base_assign_module("v20");
if (NULL == pmix_globals.mypeer->nptr->compat.bfrops) {
return PMIX_ERR_INIT;
}
} else {
/* not us */
return PMIX_ERR_NOT_SUPPORTED;
}
/* the server will be using the same bfrops as us */
pmix_client_globals.myserver->nptr->compat.bfrops = pmix_globals.mypeer->nptr->compat.bfrops;
/* mark that we are using the V2 protocol */
pmix_globals.mypeer->protocol = PMIX_PROTOCOL_V2;
/* the URI consists of the following elements:
* - server nspace.rank
* - ptl rendezvous URI
*/
uri = pmix_argv_split(evar, ';');
if (2 != pmix_argv_count(uri)) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
/* set the server nspace */
p = uri[0];
if (NULL == (p2 = strchr(p, '.'))) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
*p2 = '\0';
++p2;
nspace = strdup(p);
rank = strtoull(p2, NULL, 10);
/* save the URI, but do not overwrite what we may have received from
* the info-key directives */
if (NULL == mca_ptl_tcp_component.super.uri) {
mca_ptl_tcp_component.super.uri = strdup(uri[1]);
}
pmix_argv_free(uri);
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:client attempt connect to %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS != (rc = try_connect(&sd))) {
free(nspace);
return rc;
}
goto complete;
}
/* get here if we are a tool - check any provided directives
* to see where they want us to connect to */
if (NULL != info) {
for (n=0; n < ninfo; n++) {
if (0 == strcmp(info[n].key, PMIX_CONNECT_TO_SYSTEM)) {
system_level_only = PMIX_INFO_TRUE(&info[n]);
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_SYSTEM_FIRST)) {
/* try the system-level */
system_level = PMIX_INFO_TRUE(&info[n]);
} else if (0 == strcmp(info[n].key, PMIX_SERVER_PIDINFO)) {
pid = info[n].value.data.pid;
} else if (0 == strcmp(info[n].key, PMIX_SERVER_URI)) {
if (NULL == mca_ptl_tcp_component.super.uri) {
free(mca_ptl_tcp_component.super.uri);
}
mca_ptl_tcp_component.super.uri = strdup(info[n].value.data.string);
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_RETRY_DELAY)) {
mca_ptl_tcp_component.wait_to_connect = info[n].value.data.uint32;
} else if (0 == strcmp(info[n].key, PMIX_CONNECT_MAX_RETRIES)) {
mca_ptl_tcp_component.max_retries = info[n].value.data.uint32;
}
}
}
/* mark that we are using the V2 protocol */
pmix_globals.mypeer->protocol = PMIX_PROTOCOL_V2;
gethostname(myhost, sizeof(myhost));
/* if we were given a URI via MCA param, then look no further */
if (NULL != mca_ptl_tcp_component.super.uri) {
/* if the string starts with "file:", then they are pointing
* us to a file we need to read to get the URI itself */
if (0 == strncmp(mca_ptl_tcp_component.super.uri, "file:", 5)) {
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool getting connection info from %s",
mca_ptl_tcp_component.super.uri);
nspace = NULL;
rc = parse_uri_file(&mca_ptl_tcp_component.super.uri[5], &suri, &nspace, &rank);
if (PMIX_SUCCESS != rc) {
return PMIX_ERR_UNREACH;
}
free(mca_ptl_tcp_component.super.uri);
mca_ptl_tcp_component.super.uri = suri;
} else {
/* we need to extract the nspace/rank of the server from the string */
p = strchr(mca_ptl_tcp_component.super.uri, ';');
if (NULL == p) {
return PMIX_ERR_BAD_PARAM;
}
*p = '\0';
p++;
suri = strdup(p); // save the uri portion
/* the '.' in the first part of the original string separates
* nspace from rank */
p = strchr(mca_ptl_tcp_component.super.uri, '.');
if (NULL == p) {
free(suri);
return PMIX_ERR_BAD_PARAM;
}
*p = '\0';
p++;
nspace = strdup(mca_ptl_tcp_component.super.uri);
rank = strtoull(p, NULL, 10);
/* now update the URI */
free(mca_ptl_tcp_component.super.uri);
mca_ptl_tcp_component.super.uri = suri;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool attempt connect using given URI %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS != (rc = try_connect(&sd))) {
if (NULL != nspace) {
free(nspace);
}
return rc;
}
goto complete;
}
/* if they gave us a pid, then look for it */
if (0 != pid) {
if (0 > asprintf(&filename, "pmix.%s.tool.%d", myhost, pid)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool searching for given session server %s",
filename);
nspace = NULL;
rc = df_search(mca_ptl_tcp_component.system_tmpdir,
filename, &sd, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS == rc) {
goto complete;
}
if (NULL != nspace) {
free(nspace);
}
/* since they gave us a specific pid and we couldn't
* connect to it, return an error */
return PMIX_ERR_UNREACH;
}
/* if they asked for system-level, we start there */
if (system_level || system_level_only) {
if (0 > asprintf(&filename, "%s/pmix.sys.%s", mca_ptl_tcp_component.system_tmpdir, myhost)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool looking for system server at %s",
filename);
/* try to read the file */
rc = parse_uri_file(filename, &suri, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS == rc) {
mca_ptl_tcp_component.super.uri = suri;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool attempt connect to system server at %s",
mca_ptl_tcp_component.super.uri);
/* go ahead and try to connect */
if (PMIX_SUCCESS == try_connect(&sd)) {
goto complete;
}
free(nspace);
}
}
/* we get here if they either didn't ask for a system-level connection,
* or they asked for it and it didn't succeed. If they _only_ wanted
* a system-level connection, then we are done */
if (system_level_only) {
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp: connecting to system failed");
return PMIX_ERR_UNREACH;
}
/* they didn't give us a pid, so we will search to see what session-level
* tools are available to this user. We will take the first connection
* that succeeds - this is based on the likelihood that there is only
* one session per user on a node */
if (0 > asprintf(&filename, "pmix.%s.tool", myhost)) {
return PMIX_ERR_NOMEM;
}
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"ptl:tcp:tool searching for session server %s",
filename);
nspace = NULL;
rc = df_search(mca_ptl_tcp_component.system_tmpdir,
filename, &sd, &nspace, &rank);
free(filename);
if (PMIX_SUCCESS != rc) {
if (NULL != nspace){
free(nspace);
}
return PMIX_ERR_UNREACH;
}
complete:
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"sock_peer_try_connect: Connection across to server succeeded");
/* do a final bozo check */
if (NULL == nspace || PMIX_RANK_WILDCARD == rank) {
if (NULL != nspace) {
free(nspace);
}
CLOSE_THE_SOCKET(sd);
return PMIX_ERR_UNREACH;
}
/* mark the connection as made */
pmix_globals.connected = true;
pmix_client_globals.myserver->sd = sd;
/* setup the server info */
if (NULL == pmix_client_globals.myserver->info) {
pmix_client_globals.myserver->info = PMIX_NEW(pmix_rank_info_t);
}
if (NULL == pmix_client_globals.myserver->nptr) {
pmix_client_globals.myserver->nptr = PMIX_NEW(pmix_nspace_t);
}
if (NULL == pmix_client_globals.myserver->nptr->nspace) {
pmix_client_globals.myserver->nptr->nspace = nspace;
} else {
free(nspace);
}
if (NULL == pmix_client_globals.myserver->info->pname.nspace) {
pmix_client_globals.myserver->info->pname.nspace = strdup(pmix_client_globals.myserver->nptr->nspace);
}
pmix_client_globals.myserver->info->pname.rank = rank;
pmix_ptl_base_set_nonblocking(sd);
/* setup recv event */
pmix_event_assign(&pmix_client_globals.myserver->recv_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_READ | EV_PERSIST,
pmix_ptl_base_recv_handler, pmix_client_globals.myserver);
pmix_client_globals.myserver->recv_ev_active = true;
PMIX_POST_OBJECT(pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver->recv_event, 0);
/* setup send event */
pmix_event_assign(&pmix_client_globals.myserver->send_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_WRITE|EV_PERSIST,
pmix_ptl_base_send_handler, pmix_client_globals.myserver);
pmix_client_globals.myserver->send_ev_active = false;
return PMIX_SUCCESS;
}
static pmix_status_t connect_to_peer(struct pmix_peer_t *peer,
pmix_info_t *info, size_t ninfo)
{
char *evar, **uri;
char *filename, *host;
FILE *fp;
char *srvr, *p, *p2;
struct sockaddr_in *in;
struct sockaddr_in6 *in6;
pmix_socklen_t len;
int sd, rc;
pmix_output_verbose(2, pmix_globals.debug_output,
"ptl:tcp: connecting to server");
/* see if the connection info is in the info array - if
* so, then that overrides all other options */
/* if I am a client, then we need to look for the appropriate
* connection info in the environment */
if (PMIX_PROC_CLIENT == pmix_globals.proc_type) {
if (NULL == (evar = getenv("PMIX_SERVER_URI2"))) {
/* not us */
return PMIX_ERR_NOT_SUPPORTED;
}
/* the URI consists of elements:
* - server nspace.rank
* - ptl rendezvous URI
*/
uri = pmix_argv_split(evar, ';');
if (2 != pmix_argv_count(uri)) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
/* set the server nspace */
p = uri[0];
if (NULL == (p2 = strchr(p, '.'))) {
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
pmix_argv_free(uri);
return PMIX_ERR_NOT_SUPPORTED;
}
*p2 = '\0';
++p2;
pmix_client_globals.myserver.info = PMIX_NEW(pmix_rank_info_t);
pmix_client_globals.myserver.info->nptr = PMIX_NEW(pmix_nspace_t);
(void)strncpy(pmix_client_globals.myserver.info->nptr->nspace, p, PMIX_MAX_NSLEN);
/* set the server rank */
pmix_client_globals.myserver.info->rank = strtoull(p2, NULL, 10);
/* save the URI, but do not overwrite what we may have received from
* the info-key directives */
if (NULL == mca_ptl_tcp_component.super.uri) {
mca_ptl_tcp_component.super.uri = strdup(uri[1]);
}
pmix_argv_free(uri);
} else if (PMIX_PROC_TOOL == pmix_globals.proc_type) {
/* if we already have a URI, then look no further */
if (NULL == mca_ptl_tcp_component.super.uri) {
/* we have to discover the connection info,
* if possible. Start by looking for the connection
* info in the expected place - if the server supports
* tool connections via TCP, then there will be a
* "contact.txt" file under the system tmpdir */
filename = pmix_os_path(false, mca_ptl_tcp_component.tmpdir, "pmix-contact.txt", NULL);
if (NULL == filename) {
return PMIX_ERR_NOMEM;
}
fp = fopen(filename, "r");
if (NULL == fp) {
/* if we cannot open the file, then the server must not
* be configured to support tool connections - so abort */
free(filename);
return PMIX_ERR_UNREACH;
}
free(filename);
/* get the URI */
srvr = pmix_getline(fp);
if (NULL == srvr) {
PMIX_ERROR_LOG(PMIX_ERR_FILE_READ_FAILURE);
fclose(fp);
return PMIX_ERR_UNREACH;
}
fclose(fp);
/* up to the first ';' is the server nspace/rank */
if (NULL == (p = strchr(srvr, ';'))) {
/* malformed */
free(srvr);
return PMIX_ERR_UNREACH;
}
*p = '\0';
++p; // move past the semicolon
/* the nspace is the section up to the '.' */
if (NULL == (p2 = strchr(srvr, '.'))) {
/* malformed */
free(srvr);
return PMIX_ERR_UNREACH;
}
*p2 = '\0';
++p2;
/* set the server nspace */
pmix_client_globals.myserver.info = PMIX_NEW(pmix_rank_info_t);
pmix_client_globals.myserver.info->nptr = PMIX_NEW(pmix_nspace_t);
(void)strncpy(pmix_client_globals.myserver.info->nptr->nspace, srvr, PMIX_MAX_NSLEN);
pmix_client_globals.myserver.info->rank = strtoull(p2, NULL, 10);
/* now parse the uti itself */
mca_ptl_tcp_component.super.uri = strdup(p);
free(srvr);
}
}
/* mark that we are the active module for this server */
pmix_client_globals.myserver.compat.ptl = &pmix_ptl_tcp_module;
/* setup the path to the daemon rendezvous point */
memset(&mca_ptl_tcp_component.connection, 0, sizeof(struct sockaddr_storage));
if (0 == strncmp(mca_ptl_tcp_component.super.uri, "tcp4", 4)) {
/* separate the IP address from the port */
p = strdup(mca_ptl_tcp_component.super.uri);
if (NULL == p) {
return PMIX_ERR_NOMEM;
}
p2 = strchr(&p[7], ':');
if (NULL == p2) {
free(p);
return PMIX_ERR_BAD_PARAM;
}
*p2 = '\0';
++p2;
host = &p[7];
/* load the address */
in = (struct sockaddr_in*)&mca_ptl_tcp_component.connection;
in->sin_family = AF_INET;
in->sin_addr.s_addr = inet_addr(host);
if (in->sin_addr.s_addr == INADDR_NONE) {
free(p);
return PMIX_ERR_BAD_PARAM;
}
in->sin_port = htons(atoi(p2));
len = sizeof(struct sockaddr_in);
} else {
/* separate the IP address from the port */
p = strdup(mca_ptl_tcp_component.super.uri);
if (NULL == p) {
return PMIX_ERR_NOMEM;
}
p2 = strchr(&p[7], ':');
if (NULL == p2) {
free(p);
return PMIX_ERR_BAD_PARAM;
}
*p2 = '\0';
if (']' == p[strlen(p)-1]) {
p[strlen(p)-1] = '\0';
}
if ('[' == p[7]) {
host = &p[8];
} else {
host = &p[7];
}
/* load the address */
in6 = (struct sockaddr_in6*)&mca_ptl_tcp_component.connection;
in6->sin6_family = AF_INET6;
if (0 == inet_pton(AF_INET6, host, (void*)&in6->sin6_addr)) {
pmix_output (0, "ptl_tcp_parse_uri: Could not convert %s\n", host);
free(p);
return PMIX_ERR_BAD_PARAM;
}
in6->sin6_port = htons(atoi(p2));
len = sizeof(struct sockaddr_in6);
}
free(p);
/* establish the connection */
if (PMIX_SUCCESS != (rc = pmix_ptl_base_connect(&mca_ptl_tcp_component.connection, len, &sd))) {
PMIX_ERROR_LOG(rc);
return rc;
}
pmix_client_globals.myserver.sd = sd;
/* send our identity and any authentication credentials to the server */
if (PMIX_SUCCESS != (rc = send_connect_ack(sd))) {
PMIX_ERROR_LOG(rc);
CLOSE_THE_SOCKET(sd);
return rc;
}
/* do whatever handshake is required */
if (PMIX_SUCCESS != (rc = recv_connect_ack(sd))) {
PMIX_ERROR_LOG(rc);
CLOSE_THE_SOCKET(sd);
return rc;
}
pmix_output_verbose(2, pmix_globals.debug_output,
"sock_peer_try_connect: Connection across to server succeeded");
/* mark the connection as made */
pmix_globals.connected = true;
pmix_ptl_base_set_nonblocking(sd);
/* setup recv event */
pmix_event_assign(&pmix_client_globals.myserver.recv_event,
pmix_globals.evbase,
pmix_client_globals.myserver.sd,
EV_READ | EV_PERSIST,
pmix_ptl_base_recv_handler, &pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver.recv_event, 0);
pmix_client_globals.myserver.recv_ev_active = true;
/* setup send event */
pmix_event_assign(&pmix_client_globals.myserver.send_event,
pmix_globals.evbase,
pmix_client_globals.myserver.sd,
EV_WRITE|EV_PERSIST,
pmix_ptl_base_send_handler, &pmix_client_globals.myserver);
pmix_client_globals.myserver.send_ev_active = false;
return PMIX_SUCCESS;
}
static pmix_status_t connect_to_peer(struct pmix_peer_t *peer,
pmix_info_t *info, size_t ninfo)
{
struct sockaddr_un *address;
char *evar, **uri;
pmix_status_t rc;
int sd;
pmix_socklen_t len;
/* if we are not a client, there is nothing we can do */
if (!PMIX_PROC_IS_CLIENT) {
return PMIX_ERR_NOT_SUPPORTED;
}
/* if we don't have a path to the daemon rendezvous point,
* then we need to return an error */
if (NULL == (evar = getenv("PMIX_SERVER_URI"))) {
/* let the caller know that the server isn't available */
return PMIX_ERR_SERVER_NOT_AVAIL;
}
uri = pmix_argv_split(evar, ':');
if (3 != pmix_argv_count(uri)) {
pmix_argv_free(uri);
return PMIX_ERROR;
}
/* set the server nspace */
if (NULL == pmix_client_globals.myserver->info) {
pmix_client_globals.myserver->info = PMIX_NEW(pmix_rank_info_t);
}
if (NULL == pmix_client_globals.myserver->nptr) {
pmix_client_globals.myserver->nptr = PMIX_NEW(pmix_nspace_t);
}
if (NULL == pmix_client_globals.myserver->nptr->nspace) {
pmix_client_globals.myserver->nptr->nspace = strdup(uri[0]);
}
if (NULL == pmix_client_globals.myserver->info->pname.nspace) {
pmix_client_globals.myserver->info->pname.nspace = strdup(uri[0]);
}
/* set the server rank */
pmix_client_globals.myserver->info->pname.rank = strtoull(uri[1], NULL, 10);
/* setup the path to the daemon rendezvous point */
memset(&mca_ptl_usock_component.connection, 0, sizeof(struct sockaddr_storage));
address = (struct sockaddr_un*)&mca_ptl_usock_component.connection;
address->sun_family = AF_UNIX;
snprintf(address->sun_path, sizeof(address->sun_path)-1, "%s", uri[2]);
/* if the rendezvous file doesn't exist, that's an error */
if (0 != access(uri[2], R_OK)) {
pmix_argv_free(uri);
return PMIX_ERR_NOT_FOUND;
}
pmix_argv_free(uri);
/* establish the connection */
len = sizeof(struct sockaddr_un);
if (PMIX_SUCCESS != (rc = pmix_ptl_base_connect(&mca_ptl_usock_component.connection, len, &sd))) {
PMIX_ERROR_LOG(rc);
return rc;
}
pmix_client_globals.myserver->sd = sd;
/* send our identity and any authentication credentials to the server */
if (PMIX_SUCCESS != (rc = send_connect_ack(sd))) {
CLOSE_THE_SOCKET(sd);
return rc;
}
/* do whatever handshake is required */
if (PMIX_SUCCESS != (rc = recv_connect_ack(sd))) {
CLOSE_THE_SOCKET(sd);
return rc;
}
pmix_output_verbose(2, pmix_globals.debug_output,
"sock_peer_try_connect: Connection across to server succeeded");
/* mark the connection as made */
pmix_globals.connected = true;
pmix_ptl_base_set_nonblocking(sd);
/* setup recv event */
pmix_event_assign(&pmix_client_globals.myserver->recv_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_READ | EV_PERSIST,
pmix_ptl_base_recv_handler, &pmix_client_globals.myserver);
pmix_event_add(&pmix_client_globals.myserver->recv_event, 0);
pmix_client_globals.myserver->recv_ev_active = true;
/* setup send event */
pmix_event_assign(&pmix_client_globals.myserver->send_event,
pmix_globals.evbase,
pmix_client_globals.myserver->sd,
EV_WRITE|EV_PERSIST,
pmix_ptl_base_send_handler, &pmix_client_globals.myserver);
pmix_client_globals.myserver->send_ev_active = false;
return PMIX_SUCCESS;
}
