带有 select() 的 recvfrom() 上的 UNIX 域数据报套接字块

Question

I am using UNIX domain datagram sockets to send records from multiple clients to a single server in a multithreaded program. Everything is done within one process; I'm sending records from multiple threads to a single thread that acts as the server. All threads are assigned to separate cores using their affinity masks.

My problem is when I use select() to retrieve records from client sockets that have records in the socket buffer. I am using the same basic setup I used with a single client socket (and it worked in that context), but now it hangs (apparently it blocks) when I call recvfrom. That's surprising because the select() function has already identified the socket as available for reading.

int select_clientsockets(int64_t srvrfd, int64_t * claddr, int fds_array[], int fd_count, void * recvbuf){

    int fds_ready;
    int abc;
    int64_t cli_addr;

    FD_ZERO(&fdset);
    FD_SET(0,&fdset);

    socklen_t * len = (socklen_t * ) sizeof(struct sockaddr_un);

    fds_ready = select(3, &fdset, NULL, NULL, 0);

    for (int i = 0; i < fd_count; i++){

        fds_array[i] = 0;

        if (FD_ISSET(i, &fdset)) {
            fds_array[i] = 1;
            cli_addr = claddr[i];
            server_receive(srvrfd, recvbuf, 720, cli_addr);}
    }

    return 0;
}

The select function calls server_receive on clients where select says data are available:

int64_t server_receive(int64_t sfd, void * buf, int64_t msgLen, int64_t claddr)
{
    socklen_t * len = (socklen_t * ) sizeof(struct sockaddr_un);
    int numBytes = recvfrom(sfd, buf, BUF_SIZE, 0, (struct sockaddr *) claddr, len);

    if (numBytes == -1)
        return 0;

    return numBytes;
}

The client socket address is taken from the 3-element array "claddr" (for 3 client sockets) where the corresponding position for each client socket is filled in when the socket is created. At socket creation I also call FD_SET to set the client address into the fd_set. I think I should get the client socket address from fd_set instead, BUT they're both the same pointer value so I don't know why that would make a difference. For internet domain datagram sockets we can use getpeername() but I don't know if there is an analogous function for UNIX domain sockets -- or even if that's the problem.

Thanks very much for any help with this.

UPDATE:

Client fds are added to the global fdset struct on socket creation:

int64_t * create_socket_client(struct sockaddr_un claddr, int64_t retvals[])
{
    int sfd, j;
    size_t msgLen;
    ssize_t numBytes;
    char resp[BUF_SIZE];

    retvals[0] = 0;
    retvals[1] = 0;

    sfd = socket(AF_UNIX, SOCK_DGRAM, 0);
    if (sfd == -1)
        return retvals;

    memset(&claddr, 0, sizeof(struct sockaddr_un));
    claddr.sun_family = AF_UNIX;
    snprintf(claddr.sun_path, sizeof(claddr.sun_path),  "/tmp/ud_ucase_cl.%ld", (long) getpid());

    FD_SET(sfd,&fdset);

    retvals[0] = sfd;
    retvals[1] = (int64_t)&claddr;

     return retvals;
}

Answer 1

FD_ZERO(&fdset);
FD_SET(0,&fdset);

socklen_t * len = (socklen_t * ) sizeof(struct sockaddr_un);

fds_ready = select(3, &fdset, NULL, NULL, 0);

for (int i = 0; i < fd_count; i++){
    fds_array[i] = 0;

    if (FD_ISSET(i, &fdset)) {

Your code empties fdset then adds only 0 to fdset . So when you call select and pass it fdset , you are asking it only to check socket 0 for readiness.

You later check if sockets 0 to one less than fd_count are in fdset , but only zero could possibly be because it's the only one you asked about.

Where is the list of sockets you want to check for readiness?