Summary

Add a new IP_DONTFRAGMENT SocketOption to jdk.net.ExtendedSocketOptions.

Problem

We need to support the various socket options relating to the IPv4 DF (Do not Fragment) flag. While IPv6 does not allow packet fragmentation, there is an equivalent socket option which disables the end to end fragmentation that IPv6 can use.

This is required to support the PL PMTU discovery protocol used in QUIC.

The socket option can be a jdk.net.ExtendedSocketOption

Solution

An alternative would be to add it to java.net.StandardSocketOptions, but there is a bit too much variation in platform specific behavior of the option and it is definitely a niche requirement.

Specification

Add the following socket option to jdk.net.ExtendedSocketOptions

  /**
   * Disable IP packet fragmentation.
   *
   * <p> The value of this socket option is a {@code Boolean} that represents
   * whether the option is enabled or disabled. When {@code true} fragmentation
   * of outgoing IPv4 and IPv6 packets does not occur. This option can only be used
   * with datagram sockets. When set, care must be taken to limit outgoing packet
   * sizes to the {@link java.net.NetworkInterface#getMTU() local MTU}. Depending
   * on the implementation and the network interface, packets larger than the MTU
   * may be sent or dropped silently or dropped with an exception thrown.
   *
   * @apiNote
   * For IPv4 this option sets the DF (Do not Fragment) flag in the IP packet
   * header. This instructs intermediate routers to not fragment the packet.
   * IPv6 routers never fragment packets. Instead, fragmentation is handled
   * by the sending and receiving nodes exclusively. Setting this option for
   * an IPv6 socket ensures that packets to be sent are never fragmented, in
   * which case, the local network MTU must be observed.
   */
  public static final SocketOption<Boolean> IP_DONTFRAGMENT