Beispiel #1
0
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;
}
Beispiel #2
0
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;
}
Beispiel #3
0
static pmix_peer_t* find_peer(const pmix_proc_t *proc)
{
    pmix_status_t rc;
    pmix_peer_t *peer;
    pmix_proc_t wildcard;
    pmix_value_t *value;
    int i;

    if (NULL == proc ) {
        return pmix_globals.mypeer;
    }

    /* if the target is someone in my nspace, then use my own peer */
    if (0 == strncmp(proc->nspace, pmix_globals.myid.nspace, PMIX_MAX_NSLEN)) {
        return pmix_globals.mypeer;
    }

    if (PMIX_PROC_IS_SERVER(pmix_globals.mypeer)) {
        /* see if we know this proc */
        for (i=0; i < pmix_server_globals.clients.size; i++) {
            if (NULL != (peer = (pmix_peer_t*)pmix_pointer_array_get_item(&pmix_server_globals.clients, i))) {
                continue;
            }
            if (0 == strncmp(proc->nspace, peer->nptr->nspace, PMIX_MAX_NSLEN)) {
                return peer;
            }
        }
        /* didn't find it, so try to get the library version of the target
         * from the host - the result will be cached, so we will only have
         * to retrieve it once */
        (void)strncpy(wildcard.nspace, proc->nspace, PMIX_MAX_NSLEN);
        wildcard.rank = PMIX_RANK_WILDCARD;
        if (PMIX_SUCCESS != (rc = PMIx_Get(&wildcard, PMIX_BFROPS_MODULE, NULL, 0, &value))) {
            /* couldn't get it - nothing we can do */
            return NULL;
        }
        /* setup a peer for this nspace */
        peer = PMIX_NEW(pmix_peer_t);
        if (NULL == peer) {
            PMIX_RELEASE(value);
            return NULL;
        }
        peer->nptr = PMIX_NEW(pmix_nspace_t);
        if (NULL == peer->nptr) {
            PMIX_RELEASE(peer);
            PMIX_RELEASE(value);
            return NULL;
        }
        peer->nptr->nspace = strdup(proc->nspace);
        /* assign a module to it based on the returned version */
        peer->nptr->compat.bfrops = pmix_bfrops_base_assign_module(value->data.string);
        PMIX_RELEASE(value);
        if (NULL == peer->nptr->compat.bfrops) {
            PMIX_RELEASE(peer);
            return NULL;
        }
        /* cache the peer object */
        pmix_pointer_array_add(&pmix_server_globals.clients, peer);
        return peer;
    }

    // we are a client or tool

    /* If the target is for the server, then
     * pack it using that peer. */
    if (0 == strncmp(proc->nspace, pmix_client_globals.myserver->info->pname.nspace, PMIX_MAX_NSLEN)) {
        return pmix_client_globals.myserver;
    }

    /* if the target is another member of my nspace, then
     * they must be using the same version */
    if (0 == strncmp(proc->nspace, pmix_globals.myid.nspace, PMIX_MAX_NSLEN)) {
        return pmix_globals.mypeer;
    }

    /* try to get the library version of this peer - the result will be
     * cached, so we will only have to retrieve it once */
    (void)strncpy(wildcard.nspace, proc->nspace, PMIX_MAX_NSLEN);
    wildcard.rank = PMIX_RANK_WILDCARD;
    if (PMIX_SUCCESS != (rc = PMIx_Get(&wildcard, PMIX_BFROPS_MODULE, NULL, 0, &value))) {
        /* couldn't get it - nothing we can do */
        return NULL;
    }
    /* setup a peer for this nspace */
    peer = PMIX_NEW(pmix_peer_t);
    if (NULL == peer) {
        PMIX_RELEASE(value);
        return NULL;
    }
    peer->nptr = PMIX_NEW(pmix_nspace_t);
    if (NULL == peer->nptr) {
        PMIX_RELEASE(peer);
        PMIX_RELEASE(value);
        return NULL;
    }
    peer->nptr->nspace = strdup(proc->nspace);
    /* assign a module to it based on the returned version */
    peer->nptr->compat.bfrops = pmix_bfrops_base_assign_module(value->data.string);
    PMIX_RELEASE(value);
    if (NULL == peer->nptr->compat.bfrops) {
        PMIX_RELEASE(peer);
        return NULL;
    }
    /* need to cache the peer someplace so we can clean it
     * up later */
    return peer;
}