Socket keepalive not working

c++, linux, sockets, tcp

Solution

int val = 1;
setsockopt(socket, SOL_SOCKET, SO_KEEPALIVE, &val, sizeof val)

Just enables keepalives. You will get the default timers for keepalive probes, which you can view with the command:

sysctl net.ipv4.tcp_keepalive_time

Normally the default is couple of hours.

If you want to change the default timers, you could use this:

struct KeepConfig cfg = { 60, 5, 5};
set_tcp_keepalive_cfg(fd, &cfg);

With the helper functions here:

struct KeepConfig {
    /** The time (in seconds) the connection needs to remain 
     * idle before TCP starts sending keepalive probes (TCP_KEEPIDLE socket option)
     */
    int keepidle;
    /** The maximum number of keepalive probes TCP should 
     * send before dropping the connection. (TCP_KEEPCNT socket option)
     */
    int keepcnt;

    /** The time (in seconds) between individual keepalive probes.
     *  (TCP_KEEPINTVL socket option)
     */
    int keepintvl;
};

/**
* enable TCP keepalive on the socket
* @param fd file descriptor
* @return 0 on success -1 on failure
*/
int set_tcp_keepalive(int sockfd)
{
    int optval = 1;

    return setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &optval, sizeof(optval));
}

/** Set the keepalive options on the socket
* This also enables TCP keepalive on the socket
*
* @param fd file descriptor
* @param fd file descriptor
* @return 0 on success -1 on failure
*/
int set_tcp_keepalive_cfg(int sockfd, const struct KeepConfig *cfg)
{
    int rc;

    //first turn on keepalive
    rc = set_tcp_keepalive(sockfd);
    if (rc != 0) {
        return rc;
    }

    //set the keepalive options
    rc = setsockopt(sockfd, IPPROTO_TCP, TCP_KEEPCNT, &cfg->keepcnt, sizeof cfg->keepcnt);
    if (rc != 0) {
        return rc;
    }

    rc = setsockopt(sockfd, IPPROTO_TCP, TCP_KEEPIDLE, &cfg->keepidle, sizeof cfg->keepidle);
    if (rc != 0) {
        return rc;
    }

    rc = setsockopt(sockfd, IPPROTO_TCP, TCP_KEEPINTVL, &cfg->keepintvl, sizeof cfg->keepintvl);
    if (rc != 0) {
        return rc;
    }

    return 0;
}

Problem

I have a client connect() to a server, and when idle, it times out after a couple hours. I added setsockopt(socket, SOL_SOCKET, SO_KEEPALIVE...) with 1 sec but it didnt make a difference. Any clues on why keepalive wouldnt work? Would it make a difference if I used SOL_TCP instead of SOL_SOCKET? This is on Linux.

Original source