0

Struggling past two days with the following questions: do the loopbacked packets go through access point back to sender if the sender is subscribed to the multicast group? Even if it's not the case, is it possible to force loopback through the AP?

Also, why isn't loopback working with the following code?

char * server_addr_name = "239.255.0.1"; // multicast group
int port_number = 8888;
int enable_loopback = 1;
int udp_socket;
struct message msg; // random message
char buffer[BUFFER_SIZE];

/* create socket */
struct sockaddr_in server_addr, rcv_addr;
socklen_t server_addr_size = sizeof(struct sockaddr_in);
socklen_t rcv_addr_size = sizeof(struct sockaddr_in);

/* initialize socket */
memset(& server_addr, 0, sizeof(struct sockaddr_in));
server_addr.sin_family = AF_INET;
server_addr.sin_addr.s_addr = inet_addr(server_addr_name);
server_addr.sin_port = htons(port_number);
if ((udp_socket = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP)) == -1) {
    fprintf(stderr, "Error initializing UDP socket.\n");
    exit(EXIT_FAILURE);
}

struct ip_mreq mreq;
mreq.imr_multiaddr.s_addr = inet_addr(server_addr_name);
mreq.imr_interface.s_addr = htonl(INADDR_ANY);
if (setsockopt(udp_socket, IPPROTO_IP, IP_ADD_MEMBERSHIP, & mreq, sizeof(mreq)) < 0) {
    fprintf(stderr, "Error on setting multicast membership on socket.\n");
    exit(EXIT_FAILURE);
}
unsigned char do_enable = (unsigned char) enable_loopback;
if (setsockopt(udp_socket, IPPROTO_IP, IP_MULTICAST_LOOP,
               & do_enable, sizeof(do_enable)) < 0) {
    fprintf(stderr, "Error on setting multicast loopback on socket.\n");
    exit(EXIT_FAILURE);
}

int read_size = 0;
while (1) {

    /* send the packet */
    if (sendto(udp_socket, &msg, sizeof(struct message), 0,
            (struct sockaddr *) & server_addr, server_addr_size) == -1) {
        fprintf(stderr, "Error on sending UDP packet.\n");
    }
    else
        printf("Sent my message.\n");

    /* get response from the server/multicast address */
    read_size = recvfrom(udp_socket, buffer, BUFFER_SIZE, 0,
                            (struct sockaddr *) & rcv_addr, & rcv_addr_size);
    if (read_size < 1)
        break;
    else
        printf("Got my packet!\n");
}

Any help appreciated.

user330293
  • 1
  • 1
  • 2
  • how are you testing? are you in a test environment where sender and receiver are in the same machine? If so, I would suggest trying with 127.0.0.1 as IP and modify your hosts file adding a line: 127.0.0.1 hostname , where hostname is the computername. I have had sometimes this issues. – Picarus Sep 19 '14 at 08:27
  • @Picarus That only makes everything worse. – user207421 Sep 19 '14 at 09:37
  • @EJP, my point is first, debugging the problem to see if it is a network issue, and second, conveying that sometimes the problem is not in the code but in the system configuration. Why do you think it could be worse? It may solve the problem or not, but it can help to bring light – Picarus Sep 20 '14 at 10:05
  • Loopback works locally, it has nothing to do with AP. Logically, AP can't send packets back - if it did you would receive same packet twice. – wick Jun 26 '17 at 22:12
  • @Picarus Investigating loopback when both client and server is futile, as you can't distinguish between a proper message reception and a loopback. It certainly won't 'solve the problem'. – user207421 Aug 17 '17 at 04:48

3 Answers3

1

The argument to setsockopt() should be an int, not an unsigned char.

user207421
  • 305,947
  • 44
  • 307
  • 483
  • Well, it still didn't fix my problem. No sign of loopback at all. – user330293 Sep 19 '14 at 06:38
  • unsigned char do_enable -> int do_enable – user330293 Sep 19 '14 at 14:11
  • Although linux man page (http://linux.die.net/man/2/setsockopt) says that the argument must be socklen_t which is typedef for unsigned int. I don't think the mistake is in setsockopt() because getsockopt() shows loopback is enabled (it should enabled by default, I wrote this line of code just be sure). – user330293 Sep 19 '14 at 14:17
1

Just struggled with the same problem. Linux will not mirror the packet on the same socket even if it matches the address:port. You should create another socket udp_socket_receiver the same way as your udp_socket and recvfrom will work. Looks like IP_MULTICAST_LOOP means to loop back multicast packets locally on all listeners except the sender socket.

Janis Dzerve
  • 11
  • 1
  • 1
    you should share your thoughts by listing the to dos – Saurabh Jhunjhunwala May 11 '15 at 13:00
  • Tthis is wrong. Linux (i'm on 4.4.0) mirrors multicast datagrams to sending socket by default. You just need make sure the socket is bound to the multicast sockaddr, subscribed to the group (use `ip maddr` to check), and NOT connected. – wick Jun 26 '17 at 22:03
0

do the loopbacked packets go through access point back to sender if the sender is subscribed to the multicast group?

I don't really know for sure, but at first glance the answer is a fat No. If the gateway bounces back the packet to where it came from, you would be duplicating traffic, and that defeats the point of multicast.

However, the sender will still receive the message because anyone in the local network subscribed to the group will receive the message, and the sender is subscribed.

Even if it's not the case, is it possible to force loopback through the AP?

I don't know, but you likely don't need that. See below.

why isn't loopback working with the following code?

Because you forgot to bind the socket. You're not telling the socket where is it supposed to listen, so even though the multicast might well be bouncing properly, UDP is dropping the packet because it's attached to a random port, which is likely not 8888.

Add the following chunk between the socket() and the setsockopt(IP_ADD_MEMBERSHIP) one. Works for me:

struct sockaddr_in src;
memset(&src, 0, sizeof(src));
src.sin_family = AF_INET;
src.sin_port = htons(port_number);
src.sin_addr.s_addr = INADDR_ANY;
if (bind(udp_socket, (struct sockaddr*)&src, sizeof(src))) {
    perror("bind() failed");
    close(udp_socket);
    return EXIT_FAILURE;
}

You should also probably be close()ing the socket whenever you're done with it.

Yd Ahhrk
  • 1,088
  • 12
  • 24