Fix buffer size specification in function calls

to send(), recv() and snprintf()

ompi/communicator/comm_init.c

@@ -96,41 +96,61 @@ static void errorExit(char* msg) {
 
 // TODO: Change int return to size_t!
 static int getProcessAgentRank(uint32_t jobid, uint32_t vpid, size_t size) {
-    struct sockaddr_un strAddr;
-    socklen_t lenAddr;
-    int fdSock;
-        
-    if ((fdSock=socket(PF_UNIX, SOCK_STREAM, 0)) < 0) {
-        errorExit("socket");
+    struct addrinfo hints;
+    struct addrinfo *result, *rp;
+    const char* host_ip = "localhost";
+    const char * agent_port = getenv("DPM_AGENT_PORT");
+    if (NULL == agent_port) {
+        errorExit("Could not find DPM_AGENT_PORT env");
     }
+    memset(&hints, 0, sizeof(struct addrinfo));
+    hints.ai_family = AF_UNSPEC;
+    hints.ai_socktype = SOCK_STREAM; //opting for reliable sequenced socket type
+    hints.ai_flags = 0;
+    hints.ai_protocol = IPPROTO_TCP;
+
+    int error_code = getaddrinfo(host_ip, agent_port, &hints, &result);
+    if (0 != error_code) {
+        errorExit("Could not get addrinfo!");
+    }
+
+    int socket_fd = 0;
+    for (rp = result; rp != NULL; rp = rp->ai_next) {
+        socket_fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
+
+        if (socket_fd == -1) {
+            continue;
+        }
 
-    strAddr.sun_family=AF_LOCAL;     /* Unix domain */
-    const char * socket_path = getenv("DPM_AGENT_SOCKET"); // check for success
-    if (NULL == socket_path) {
-        errorExit("Could not find DPM_AGENT_SOCKET env");
+        if (0 == connect(socket_fd, rp->ai_addr, rp->ai_addrlen)) {
+            break; //successfully connected
+        } else {
+            close(socket_fd); //close socket when we cannot connect
+        }
     }
-    strcpy(strAddr.sun_path, socket_path);
-    lenAddr=sizeof(strAddr.sun_family)+strlen(strAddr.sun_path);
-    if (connect(fdSock, (struct sockaddr*)&strAddr, lenAddr) !=0 ) {
-        errorExit("connect");
+    freeaddrinfo(result); //free result
+    // exit if the result-linked-list has been traversed until the end without a successful
+    // connection
+    if (NULL == rp) {
+        errorExit("Could not connect to host. Host not available or host/port invalid.");
     }
 
     pid_t pid = getpid();
 
     char info_to_send[BUFFLEN];
-    snprintf(info_to_send, BUFFLEN+1,
+    snprintf(info_to_send, BUFFLEN,
              "%d,%u,%u,%zu",
              pid, vpid, jobid, size);
-    if (send(fdSock, info_to_send, BUFFLEN+1, 0) < 0) {
+    if (send(socket_fd, info_to_send, BUFFLEN, 0) < 0) {
         errorExit("send");
     }
 
     char rank_to_recv[BUFFLEN];
-    recv(fdSock, rank_to_recv, BUFFLEN+1, 0);
+    recv(socket_fd, rank_to_recv, BUFFLEN, 0);
     int received_rank = (int)strtol(rank_to_recv, NULL, 0);
     printf("Received from server: %d\n", received_rank);
 
-    close(fdSock);
+    close(socket_fd);
 
     return received_rank;
 }