@riccardv commented on this pull request.
// for Reply and TCP sends from P-CSCF server port, for Reply and UDP sends from P-CSCF client port
- src_port = dst_proto == PROTO_TCP ? s->port_ps : s->port_pc; + // Check send socket + struct socket_info * client_sock = grep_sock_info(via_host.af == AF_INET ? &ipsec_listen_addr : &ipsec_listen_addr6, src_port, dst_proto); + if(client_sock) { + // for Reply and TCP sends from P-CSCF server port, for Reply and UDP sends from P-CSCF client port + src_port = dst_proto == PROTO_TCP ? s->port_ps : s->port_pc;
- // for Reply and TCP sends to UE client port, for Reply and UDP sends to UE server port - dst_port = dst_proto == PROTO_TCP ? s->port_uc : s->port_us; + // for Reply and TCP sends to UE client port, for Reply and UDP sends to UE server port + dst_port = dst_proto == PROTO_TCP ? s->port_uc : s->port_us; + } + else + { + src_port = s->port_pc; + dst_port = s->port_us; + }
Hi @alexyosifov ,
without the patch the ports in case of TCP are always: src_port = s->port_ps dst_port = s->port_uc That are the ones used in case the user agent open the connection.
A pre check is necessary because if this socket is not already open, the use the ports: src_port = s->port_pc; dst_port = s->port_us; that are the port from Proxy Client (pc) -> User Server (us). Is a attempt to fallback to the other opened socket.
This case can be happens when INVITE transaction is very long due to long Ringing time phase.