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;
}
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 try_connect(int *sd)
{
char *p, *p2, *host;
struct sockaddr_in *in;
struct sockaddr_in6 *in6;
size_t len;
pmix_status_t rc;
bool retried = false;
pmix_output_verbose(2, pmix_ptl_base_framework.framework_output,
"pmix:tcp try connect to %s",
mca_ptl_tcp_component.super.uri);
/* mark that we are the active module for this server */
pmix_client_globals.myserver->nptr->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)) {
/* need to skip the tcp4: part */
p = strdup(&mca_ptl_tcp_component.super.uri[7]);
if (NULL == p) {
PMIX_ERROR_LOG(PMIX_ERR_NOMEM);
return PMIX_ERR_NOMEM;
}
/* separate the IP address from the port */
p2 = strchr(p, ':');
if (NULL == p2) {
free(p);
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
return PMIX_ERR_BAD_PARAM;
}
*p2 = '\0';
p2++;
host = p;
/* 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);
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
return PMIX_ERR_BAD_PARAM;
}
in->sin_port = htons(atoi(p2));
len = sizeof(struct sockaddr_in);
} else {
/* need to skip the tcp6: part */
p = strdup(&mca_ptl_tcp_component.super.uri[7]);
if (NULL == p) {
PMIX_ERROR_LOG(PMIX_ERR_NOMEM);
return PMIX_ERR_NOMEM;
}
p2 = strchr(p, ':');
if (NULL == p2) {
free(p);
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
return PMIX_ERR_BAD_PARAM;
}
*p2 = '\0';
if (']' == p[strlen(p)-1]) {
p[strlen(p)-1] = '\0';
}
if ('[' == p[0]) {
host = &p[1];
} else {
host = &p[0];
}
/* 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);
PMIX_ERROR_LOG(PMIX_ERR_BAD_PARAM);
return PMIX_ERR_BAD_PARAM;
}
in6->sin6_port = htons(atoi(p2));
len = sizeof(struct sockaddr_in6);
}
free(p);
retry:
/* establish the connection */
if (PMIX_SUCCESS != (rc = pmix_ptl_base_connect(&mca_ptl_tcp_component.connection, len, sd))) {
/* do not error log - might just be a stale connection point */
return rc;
}
/* 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))) {
CLOSE_THE_SOCKET(*sd);
if (PMIX_ERR_TEMP_UNAVAILABLE == rc) {
/* give it two tries */
if (!retried) {
retried = true;
goto retry;
}
}
PMIX_ERROR_LOG(rc);
return rc;
}
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;
}