I made working patch on the current master branch:

--- a/src/core/tcp_main.c
+++ b/src/core/tcp_main.c
@@ -4324,10 +4324,25 @@ static inline int handle_new_connect(struct socket_info *si)
        struct tcp_connection *tcpconn;
        socklen_t su_len;
        int new_sock;
+       int optval;

        /* got a connection on r */
        su_len = sizeof(su);
        new_sock = accept(si->socket, &(su.s), &su_len);
+
+       if ((optval = TICKS_TO_S(cfg_get(tcp, tcp_cfg, send_timeout)))) {
+               optval *= 1000;
+               if(setsockopt(new_sock, IPPROTO_TCP, TCP_USER_TIMEOUT, &optval,
+                                  sizeof(optval))
+                               < 0) {
+                       LM_WARN("failed to set TCP_USER_TIMEOUT: %s\n",
+                                       strerror(errno));
+               } else {
+                       LM_INFO("Set new connection TCP_USER_TIMEOUT=%d ms\n", optval);
+               }
+       }
+
+
        if(unlikely(new_sock == -1)) {
                if((errno == EAGAIN) || (errno == EWOULDBLOCK))
                        return 0;

Now the retransmits do what is configured and time out after 3s.

Jul 27 18:14:15 p01t-1 kamailio.sbc[1182787]: INFO: <core> [core/tcp_main.c:4373]: handle_new_connect(): Set new connection TCP_USER_TIMEOUT=3000 ms
...
Jul 27 18:14:49 p01t-1 kamailio.sbc[1182775]: INFO: sbc:<script>: >>> Sending Request: INVITE sip:22@192.168.41.157:51151;line=q9yhaejq;ob (10.84.65.131:5082 -> 10.84.65.130:51151: sip:193.84.65.130:51151;transport=tcp)
Jul 27 18:14:52 p01t-1 kamailio.sbc[1182785]: NOTICE: <core> [core/tcp_read.c:267]: tcp_read_data(): error reading: Connection timed out (110)

Readings:

https://stackoverflow.com/questions/5907527/application-control-of-tcp-retransmission-on-linux
https://man7.org/linux/man-pages/man7/tcp.7.html

Although said "This option, like many others, will be inherited by the socket returned by accept(2), if it was set on the listening socket." => NO, it is not inherited (I tried it after init_sock_keepalive(sock_info->socket); line in tcp_main.c). So it seems it must be done on the spawned socket in handle_new_connect().

Needs more testing and must be done more pretty.

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.