plipbox: Introduction ===================== This document gives a technical introduction on how the plipbox device operates. The plipbox device firmware implements a bridge for network packets arriving on the parallel port of the Amiga (called PLIP) and the local Ethernet connected to the plipbox hardware. 1. NAT device ------------- Version 0.1 and 0.2 of plipbox implemented a NAT device between the parallel PLIP interface and the Ethernet adapter. A NAT device connects two distinct IP networks, i.e. both the PLIP and Ethernet side of the plipbox have different IP networks and thus the plipbox has to have two addresses (one for each sub net): +-----plipbox-----+ | 1a:11:a1:a0:47:11 +-------+ IP +-----------+ +----------+ ARP/IP | Amiga | <---> | PLIP Port |-NAT-| Eth Port | <---> Local Network +-------+ p2p +-----------+ +----------+ lan 192.168.0.1 192.168.0.2 192.168.2.133 192.168.2.x magplip.device | | mask: 255.255.255.0 +-----plipbox-----+ gw: 192.168.2.1 In this setup you have to configure quite a lot: You have to set your Amiga IP and the PLIP port IP in an own sub net. Additionally, you have to set your Ethernet port IP to an address suitable for your local network. If you have a DHCP server in your local network then the plipbox firmware can use this protocol to automatically acquire the Ethernet port IP address and sub net mask. On the Amiga side you have to select a point-to-point networking setup in your TCP/IP software stack. This defines one address for the magPLIP interface (here 192.168.0.1) and one for the PLIP Port in plipbox (here 192.168.0.2). All packets sent from the Amiga will arrive at the gateway on the PLIP Port. Now NAT takes place and the sender addresses are replaced in all IP packets before sending them to the Ethernet: A packet originating from the Amiga (192.168.0.1) is sent with the Eth Port as sender (192.168.2.133) to the local network. If a reply IP packet arrives at the plipbox's Eth Port then NAT is reversed and replaces the destination address of the Eth Port (192.168.2.133) with the one of the Amiga (192.168.0.1). Then the packet is sent via PLIP to the Amiga. Note that in contrast to other NAT devices here no port remapping is done: We NAT only a single machine (the Amiga) and we don't use our own ports locally. This simplifies NATting a bit as not port maps need to be managed. On the point-to-point link only IP packets are transported. All other types especially ARP packets make no sense here as a HW address to IP mapping is not required here. Unfortunately, some Amiga TCP/IP stacks do not support p2p links very well and try to perform ARP here, too :/ However, ARP must be spoken on Eth Port to map local IP addresses to the Ethernet MAC addresses. Whenever an IP packet (after NAT) is ready to be sent out via Ethernet then we have to find out what Ethernet MAC address has the destination. If the destination IP is on the local sub net (determined with the netmask of the local network) then ARP is used to find the corresponding MAC. Otherwise we ARP the gateway (gw) IP address on the local sub net and send the packet there. To make ARPing more performant plipbox keeps an ARP cache. This is a table of IP to MAC address mappings that were already used on the local sub net. The MAC address of the local ethernet adapter on the plipbox is set by the firmware. The default firmware already ships with a pre-defined address. If you use multiple plipboxes in one local sub net then you have to alter this address for individual devices. Regarding MTU size this approach actually can apply two sizes, one for each sub net. While the Ethernet is typically set to 1500 the PLIP link is not limited to this value and can use arbitrary ones. However, if the PLIP MTU is larger than 1500 then the plipbox needs to fragment the packets but this is currently not implemented. In summary, the NAT approach of the firmware works well but needs quite a lot of configuration. Furthermore, the NATing applied causes trouble to protocols that use IP addresses inside the protocol. Most notably, the FTP needs to use passive mode in this setup to work correctly. Another cause for trouble in this setup is the NAT that is already applied in most local home networks when reaching the Internet. E.g. a DSL router NATs all outgoing traffic. So a double NAT is applied to all Amiga packets when they run through plipbox, local network, and finally the Internet. 2. MAC bridge ------------- Starting with version 0.3 plipbox the NAT approach is replaced by a MAC bridge. This bridge is more "transparent" than NATting and should overcome the problems associated with NAT. The old Amiga network device (magplip.device) that offered a point-to-point link is replaced with a new device driver (plipbox.device) that announces an Ethernet compatible device. The TCP/IP stack on the Amiga now directly generates Ethernet packets with a 1500 MTU and transfers them to the plipbox. There the frame is directly (i.e. without any alteration) passed to the Ethernet adapter and send to the local network. Also any incoming frame on the local network is passed to the Amiga plipbox device driver and deliverd as an Ethernet packet to the TCP/IP stack. +-----plipbox-----+ 1a:11:a1:a0:47:11 | 1a:11:a1:a0:47:11 +-------+ ARP/IP +-----------+ +----------+ ARP/IP | Amiga | <----> | PLIP Port |-COPY-| Eth Port | <---> Local Network +-------+ +-----------+ +----------+ lan 192.168.2.133 | | 192.168.2.x plipbox.device | | mask: 255.255.255.0 +-----plipbox-----+ gw: 192.168.2.1 Now the plipbox firmware is most transparent and essentially looks like an ethernet device directly attached to the Amiga. In this setup no NAT is applied. You now configure your Amiga's TCP/IP stack like it is directly attached to the local network. Give yourself a static IP or use DHCP, set the local network's net mask, and the default gateway. Also the stack now needs to do ARPing to map the IP addresses to Ethernet MAC addresses and it manages an own ARP cache. The plipbox firmware is now very slim and zero-config. The only essential parameter is the MAC address of the Eth Port. It has to have the same value as it is used in the plipbox.device driver. The address will be automatically transferred by the driver to the firmware on startup. By default both use the mac address 1a:11:a1:a0:47:11. The device driver uses the mac address to generate the correct source address in each frame while the plipbox firmware uses the mac address to setup a HW Ethernet filter so that the Eth Port only reports packets for this or the broadcast address. In summary, with the new approach the Amiga has a bit more to do (e.g. handling ARP packets) but all problems that were observed with NAT are gone now. EOF