Net::SIP:::SocketPool - manage sockets related to a leg


  my $pool = Net::SIP::SocketPool->new(...)
  $pool->sendto($packet, [ip,port,family], \&callback)


SocketPool manages a collection of sockets associated with a Leg. This is usually an unconnected socket (i.e. UDP or TCP listen socket) and mayby some connected sockets. While in UDP a packet can be received and sent using an unconnected socket this is not possible in TCP and therefore these connected socket have to be maintained somehow. Also, it is expected in TCP that a response will be sent back through the same TCP connection as the request came in, if possible.

SocketPool is usually not used directly but will be created when a new Leg gets created.



The constructer creates a new SocketPool for protocol PROTO (udp, tcp or tls) with FD as the master socket. If CONNECTED is true this master socket is connected and DST will in this case be interpreted as the peer of the socket. But a connected master socket makes only sense for UDP and only if the communication should be limited to specific party, like an outgoing SIP proxy. In the common case that CONNECTED is false the optional DST given as [ip, port, family] will be interpreted as restriction for the communication, i.e. it will be forced as destination in sendto no matter what was given and it will be checked that any received data origin from the expected peer DST.

With the optional TLS argument a hash can be givevn wth arguments used in creation of the IO::Socket::SSL objects when <PROTO> is tls. This typically includes location of the certificate and key with SSL_cert_file and SSL_key_file. These arguments will be used for both server and client SSL sockets which also means that the certificate configured as server certificates will also be used as client certificates if the peer requires authentication with client certificates. The special argument verify_client in TLS can be used to require authentication with client certificates by the peer. It can be set to 0 for no client certificates, -1 for optional and 1 for required client certificates.



This method is used indirectly from Leg::deliver to deliver a new packet to its destinination.

This will deliver the Net::SIP::Packet PKT to the target DST given as hash with addr, port, family and will invoke CALLBACK when done. Callback can be anything accepted by invoke_callback from Net::SIP::Util.

With TCP the SocketPool will try to find an existing connected socket to the target first before creating a new one. For response packets it will prefer the socket where the request packet came in, if possible.

With UDP instead it will just use the master socket for sending.


This will just return the FD for the master socket. This is used by Leg in case the SocketPool was created outside the Leg.

attach_eventloop(LOOP, CALLBACK)

This attaches the SocketPool to a Net::SIP::Dispatcher::EventLoop object so that it can be used for event based I/O. This attaches CALLBACK as read handler to the given LOOP to handle new packets coming in through the sockets inside the SocketPool. It will accept any callback suitable for invoke_callback and will invoke it with [PKT, FROM] where PKT is the freshly read Net::SIP::Packet and FROM the origin of this packet as hash. This hash includes addr, port of the sender, family of the socket, proto as the used protocol (i.e. 'udp', 'tcp' or 'tls') and socket for the local socket object where the packet was received on. This socket is either an IO::Socket or IO::Socket::SSL object and is only intended for passive use, for example to extract the certificate send by the peer.

If LOOP is undef it will just detach from the current loop.

This function is used from inside Net::SIP::Dispatcher to attach a legs sockets to the event loop and process incoming data.

Additionally to these methods the internal configuration can be adjusted with use or import:

    use Net::SIP::SocketPool (MAX_SIP_HEADER => 2**14, ... );

The following settings are possible this way:


maximum size of SIP header, default 2**14


maximum size of SIP body, default 2**16


This is maximum size of remembered incoming requests per socket. These requests need to be remembered so that outgoing responses can be sent back through the same connection as the request came in. This defaults to 30.


The minimal time for socket expiration and the maximum time. These default to 15 and 120 (seconds). The exact time for expiration depends on the number of sockets in the socketgroup, i.e. the more sockets the shorter the expiration timeout.


The timeout used for establishing a TCP connection. Default to 10 (seconds).


The amount of data it tries to read within a single sysread, default 2**16.