NAPT Installation and Configuration
March 16th, 2001 (Check here for the latest version.)
This document describes the installation and configuration of the Network Address and Procotol Translator (NAPT), which we will refer to as 'translator' from hereon. The purpose of the translator is to convert packets as they cross the boundary between IPv4 and IPv6 networks. It achieves this by translating the various protocol fields from one version of IP to the other. The translator behaves as a router, receiving IPv4 packets from one network and transmitting IPv6 packets on the other (and vice-versa).
This document describes the system requirements of the translator, how to build it, a brief description of its implementation, how to configure it for translation, an example scenario, and finally a list of limitations.
Our implementation should work on any version of NT that supports the AO_OPTION_IP_HDRINCL TDI option to send 'raw' IPv4 packets. This is currently available only on Windows 2000 Beta 2 and later systems. However, even in those systems it is not possible to send IPv4 fragments (i.e., IPv4 packets with the offset field set to non-zero values). This capability may become available in the future.
The translator is linked with the MSRIPv6 driver in order to receive and send IPv6 packets. It is necessary that a version of the MSRIPv6 driver is installed that exports the required symbols. For this version of the translator, this is MSRIPv6 version 1.4.
(We include a binary with this release, so you only need to build your own if you change the translator in some way)
The translator is built as a regular device driver for Windows 2000. For further instructions on how to build Windows 2000 drivers, please consult the "Building" instructions in the MSRIPv6 readme.
Most of our development has occurred on Windows NT 4.0 machines using Visual C++ 5.0, Platform SDK of Oct 1997 and the NT DDK Version 4.0, as well as the September 1998 Windows 2000 Beta 2 releases of the Platform SDK and NT DDK Version 5.0. We've also successfully compiled the translator with Visual C++ 6.0 and later releases of the SDK and DDK.
The translator is implemented as a Windows 2000 device driver. The driver is designed to be dynamically loaded and unloaded without having to restart Windows 2000.
The translator uses the MSRIPv6 driver to send and receive IPv6 packets. To receive packets it installs itself as an 'Interface' in the MSRIPv6 stack and registers routes for packets that should be forwarded to the translator for conversion from IPv6 to IPv4. It uses the MSRIPv6 IPv6Send0() function to send fully formed IPv6 packets.
To receive IPv4 packets the translator installs itself as an NDIS protocol driver and filters out those that it should translate to IPv6 packets. The code that implements this is in lan.c. (Ideally the translator would install itself as an interface to the IPv4 stack, similar to how it works with the IPv6 stack.) The translator sends IPv4 packets using TDI and setting the IPHDRINCL option. The code that implements this is in ippub.c.
The bulk of the translation code is in napt.c. In particular, the IPv4/IPv6 protocol field conversion is done in the NaptIP4toIP6() and NaptIP6toIP4() functions. Similarly, ICMPv4/ICMPv6 conversion is done in NaptIcmp4toIcmp6() and NaptIcmp6toIcmp4(). Note that these functions may call each other recursively when translating an ICMP error message.
The protocol conversion rules implemented by these functions are spelled out in the documents listed in the reference section at the end of this page. (Unfortunately, there is no RFC on how exactly the translation should occur. Update - the two referenced documents were recently approved for Proposed Standard and are now RFCs - Ed) For the most up-to-date conversion rules see sections 3 & 4 in Nordmark's SIIT Internet Draft .
Unfortunately, the translator driver (napt.sys) is not as easily installable as the MSRIPv6 stack. To install it you need to run the provided napt.reg file to install some registry keys the translator uses. You will also need to place napt.sys into your %SystemRoot%\system32\drivers directory. Once you do that, it is necessary to reboot the system. And while you will only need to edit the registry once to configure the translator (see "Configuration" below), you will need to start the driver manually using the command "net start napt" whenever you reboot the system.
The translator can perform two types of address translation: stateless and stateful. It can be configured to perform either or both of these methods. Stateless Translation
For stateless address translation, the translator maintains no information as to which IPv4 address a given IPv6 address corresponds to (and vice-versa). All the information the translator needs to convert addresses from one protocol to the other is provided in the packets themselves through the use of IPv4-mapped and IPv4-translated addresses on the IPv6 side. That is to say, an IPv4 packet will be translated into an IPv6 packet with addresses composed of a special prefix and an embedded IPv4 address. The IPv6-only hosts must have a valid IPv4-translated address (i.e., an IPv6 address with the ::ffff:0:0:0/96 prefix), and they will communicate with IPv4-only hosts by addressing them using IPv4-mapped addresses (i.e. an IPv6 address with the ::ffff:0:0/96 prefix). You can not use arbitrary IPv6 addresses with stateless translation.
The following illustration shows how addresses are translated when an IPv6-only host with an (IPv4-translated) address of ::ffff:0:18.104.22.168 is communicating with an IPv4-only host with an address of 22.214.171.124. The IPv4-only host thinks it is talking to another IPv4 host (with an address of 126.96.36.199), while the IPv6-only thinks it is talking to another IPv6 host (with an IPv4-mapped address of ::ffff:188.8.131.52). Stateless translation of a packet from IPv6 to IPv4 IPv6 Packet Source0:0:0:0:ffff:0:184.108.40.206 Destination0:0:0:0:0:ffff:220.127.116.11 ----> IPv4 Packet Source18.104.22.168 Destination22.214.171.124 Stateless translation of a packet from IPv4 to IPv6 IPv4 Packet Source126.96.36.199 Destination188.8.131.52 ----> IPv6 Packet Source0:0:0:0:0:ffff:184.108.40.206 Destination0:0:0:0:ffff:0:220.127.116.11 Stateful Translation
For stateful address translation, the translator maintains explicit mappings between arbitrary IPv4 and IPv6 addresses. The use of IPv4-mapped and IPv4-translated addresses is not required. When converting addresses, the translator simply consults its pre-configured table to determine the corresponding address to use with the other protocol.
The following illustration shows how addresses are translated when an IPv6-only host with an address of 2002:836b:4179:0:0:0:836b:4179 is communicating with an IPv4-only host with an address of 18.104.22.168. The IPv4-only host thinks it is talking to another IPv4 host (with an address of 22.214.171.124), while the IPv6-only thinks it is talking to another IPv6 host (with an address of fec0::cf2e:820e). The translator keeps an internal mapping between 2002:836b:4179:0:0:0:836b:4179 and 126.96.36.199 and between fec0::cf2e:820e and 188.8.131.52. Note that unlike the stateless translation case, the address values used on the IPv6 side to represent IPv4 nodes are completely arbitrary - there is no requirement that they be related to their corresponding IPv4 address. Stateful translation of a packet from IPv6 to IPv4 IPv6 Packet Source2002:836b:4179:0:0:0:836b:4179 Destinationfec0::cf2e:820e ----> IPv4 Packet Source184.108.40.206 Destination220.127.116.11 Stateful translation of a packet from IPv4 to IPv6 IPv4 Packet Source18.104.22.168 Destination22.214.171.124 ----> IPv6 Packet Sourcefec0::cf2e:820e Destination2002:836b:4179:0:0:0:836b:4179
For either stateless or stateful translation, in order for packet translation to occur, such packets need to be routed via a translating-router that is configured to reach the IPv4 and IPv6 networks specified in the packet. Our translator implements such a translating-router between an IPv4 and an IPv6 network. However, with our current implementation, the translator does not participate in routing protocols. Everything must be statically configured.
On the IPv4 side, routes to the translator must be established for the IPv4 address(es) or address ranges you wish it to translate. So set these routes on every IPv4 client or on the IPv4 subnet's default router. Similarly, on the IPv6 side, routes to the translating-router must be established for the IPv6 prefixes or specific host addresses it translates (the translating-router can be configured to advertise these routes so you don't have to set them manually, see next paragraph). Also, once IPv6 packets are received by the translating-router, they need to be forwarded to the translation interface proper (the translation interface will only appear once Napt is started, and will typically be the highest numbered interface). These routes are installed automatically when Napt starts based on information entered in the registry as part of the configuration procedure discussed below. However, in order for the translating-router to invoke those routes on packets received from elsewhere, the receiving interface on the IPv6 side must be marked as "forwarding". Determine the interface index of the interface on the IPv6 side of the translating-router and run "ipv6 ifc ifindex forwards". Finally, routes from the translator to the IPv4 and IPv6 networks must be added to the translating machine's routing tables. These will cause the successfully translated packets to leave the translating-router in the right direction.
As mentioned above, the translator is capable of sending out 'IPv6 Router Advertisements' with a new type of prefix option that contains a prefix that is routable via the translating-router. (Implementation note: this is accomplished by setting the ND_PREFIX_FLAG_ROUTE prefix option flag to indicate this special prefix option - see icmp6.h and neighbor.c in the MSRIPv6 sources) Currently, only MSRIPv6 machines handle this special prefix option flag. (NOTE: We have yet to write an Internet Draft for this) The translator will advertise these routes on any interface marked to do so; determine the interface index of the interface on the IPv6 side of the translating-router and run "ipv6 ifc ifindex advertises". Unfortunately, IPv4 routes to the translating-router still need to be manually configured.
Additionally, if using stateless address translation, it is necessary to manually configure IPv6 clients with their IPv4-translated address. This is done using the "ipv6 adu" command. For example, if you wish interface 3 on a particular IPv6 host to appear to the IPv4 network as if it had an IPv4 address of 126.96.36.199, then you would run "ipv6 adu 3/::ffff:0:188.8.131.52" to set that address.
There are two registry entries that are used in configuring the translator proper, one for each type of translation desired. These are:
- StatelessV4RangeFilters: a list of IPv4 address ranges which the router will translate to IPv6.
- StatefulTranslations: a list of explicit mappings between real and translated addresses.
NOTE: The values placed into these two parameters are of type REG_MULTI_SZ. It seems that only regedt32.exe supports this type. The other registry editor, regedit.exe, does not support that particular type.
Using regedt32.exe, drill down to HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\napt\Parameters (if this key is not present, make sure you ran the napt.reg file as mentioned in "Installation" above) and then add the appropriate values (use "Add Value" off of the "Edit" menu, and be sure to select a Data Type of "REG_MULTI_SZ" from the drop down box) as follows:
StatelessV4RangeFilters - REG_MULTI_SZ: place zero or more 'IPv4AddrLow IPv4AddrHigh' pairs into this registry, where each pair defines an address range that the translator will filter out from the IPv4 packets it receives and translate them to IPv6 packets. For example, for each of the following entries10.0.100.1 10.0.100.254 184.108.40.206 220.127.116.11 18.104.22.168 22.214.171.124
the translator will translate all IPv4 packets with a destination address starting at the IPv4AddrLow and ending at IPv4AddrHigh. Also, as long as there are any entries present in this registry value, the system will automatically add a route to the translator for IPv6 packets sent to addresses with the IPv4-mapped prefix. This will cause the translator to translate the appropriate IPv6 packets to IPv4 packets.
StatefulTranslations - REG_MULTI_SZ: place zero or more 'RealAddr TranslatedAddr' pairs into this registry entry, where each pair defines an explicit mapping between a 'real' address and a 'translated' address. Each machine that you wish to make accessible to the other protocol should have an entry. Put the host's native address (either IPv4 or IPv6) first, followed by the address (either IPv4 or IPv6, whichever one the native one isn't) that you wish it to appear as to the other protocol. For example,126.96.36.199 beef:feed::1234:5678 3ffe:1cff::bead:ed:cafe:deed 188.8.131.52
assigns the (translated) IPv6 address of beef:feed:1234:5678 to the host with the (real) IPv4 address of 184.108.40.206. And likewise, it assigns the IPv4 address of 220.127.116.11 to the host with the IPv6 address of 3ffe:1cff::bead:ed:cafe:deed. Put as many entries as you need. The system will add IPv4 filters or IPv6 routes for the translated addresses so that when packets with these addresses arrive at the translating-router, they will undergo translation.
Once you have configured everything, you can start the translator operating via "net start napt" (you can stop the translator at any time via "net stop napt").
Here is a simple example scenario of stateless translation between two subnets. (The IPv4 addresses used here are for example purposes, you'll want to substitute addresses from your own allocation) On the IPv4 side, we have a subnet with address 172.31.42.0/24. Say we want to set up the translator to make the IPv6 side appear to IPv4 side as if it was the IPv6 subnet with address 172.31.43.0/24. And since we're doing stateless translation, the IPv6 subnet address is constrained to be under the IPv4-translated prefix. Combining these two restrictions gives us an IPv6 subnet of ::ffff:0:172.31.43.0/120. To keep this simple, let's consider three specific hosts: GACK as an IPv4 host, GADZOOKS as the IPv6/IPv4 translator, and YOWZA as an IPv6 host. GACK has an address of 172.31.42.50. GADZOOKS has two interfaces; the one on the IPv4 side has an address of 172.31.42.56, while the one on the IPv6 side has a link-local address of fe80::2a0:ccff:fe39:7dc5. Finally, YOWZA has an address of ::ffff:0:172.31.43.212, which makes it appear to the IPv4 world as if it were 172.31.43.212. The figure below depicts this scenario. Example stateless address translation setup IPv4 Host [GACK] 172.31.42.50 | IPv4 subnet 172.31.42.0/24 172.31.42.56 Translator [GADZOOKS] fe80::2a0:ccff:fe39:7dc5 | IPv6 subnet ::ffff:0:172.31.43.0/120 ::ffff:0:172.31.43.212 IPv6 Host [YOWZA] How to set this up
Let's start with the client hosts. On the IPv4 side, GACK needs to have its IP address of 172.31.42.50 assigned, which you accomplish via one of the usual ways. It also needs to have a route for 172.31.43.0/24 pointing at GADZOOKS. For a Windows NT/2000 machine, you would set this up via "route add 172.31.43.0 MASK 255.255.255.0 172.31.42.56". On the IPv6 side, YOWZA needs to have its IPv4-translated address assigned, you do this via "ipv6 adu ifindex/::ffff:0:172.31.43.212" (substitute the appropriate interface number for "ifindex"). YOWZA also needs a route for the IPv4-mapped prefix pointing at GADZOOKS. However, you shouldn't have to configure this manually, as YOWZA should pick this up from router advertisements sent by GADZOOKS.
As the translator, GADZOOKS must be configured to 1) accept for translation all IPv4 packets sent to the 172.31.43.0/24 subnet, 2) have a route for the IPv4-translated prefix out the interface on the IPv6 side of the translator, 3) advertise a route for the IPv4-mapped prefix on the IPv6 side, 4) have a route for the IPv4-mapped prefix to the special translation interface on the router, and 5) have a route for the 172.31.42.0/24 prefix out the interface on the IPv4 side of the translator.
- This is done via the StatelessV4RangeFilters registry entry as described in "Configuration" above. For this example, you would set the value to the single line "172.31.43.1 172.31.43.254".
- Identify the interface that is on the IPv6 side of the translator. Then using the index of that interface, run "ipv6 rtu ::ffff:0:0:0/96 ifindex" to direct all packets for IPv4-translated addresses out the IPv6 side of the translator.
- Mark this same interface as both forwarding and advertising via "ipv6 ifc ifindex forwards advertises".
- When Napt is started, it will automatically install a published route for the IPv4-mapped prefix to the translator as a side-effect of step (1), so you don't have to do anything here. (This will show up in the output of "ipv6 rt" as something like "::ffff:0:0/96 -> translatorifindex/fe80::1")
- If you can ping the IPv4 hosts from the translator, you already such a route in place. Otherwise, use the regular IPv4 commands to create one, something like "route add 172.31.42.0 MASK 255.255.255.0 172.31.42.56" should do it.
With configuration complete, starting the translator ("net start napt") will allow packets to flow from one subnet to the other.
An IPv4 packet sent to 172.31.43.212 by GACK will be routed to GADZOOKS, which will translate it from IPv4 to IPv6 and send it to YOWZA. The resulting IPv6 packet will have a IPv4-translated address as its destination (::ffff:0:172.31.43.212) and a IPv4-mapped address (::ffff:172.31.42.50) as its source.
An IPv6 packet sent to ::ffff:172.31.42.50 (the IPv6-mapped address corresponding to GACK) by YOWZA will be sent to the translator (due to the route for the IPv6-mapped prefix pointing at fe80::2a0:ccff:fe39:7dc5 on that subnet), then routed internal to the router to the translation interface (due to the route installed automatically by the translator at startup), which will translate it from IPv6 to IPv4 and send it to GACK. The resulting IPv4 packet will have a destination address of 172.31.42.50 and a source address of 172.31.43.212.
On GACK, try "ping 172.31.43.212". From YOWZA, try "ping6 ::ffff:172.31.42.50". To really stress things, start "ttcp -r -P4" going on GACK, and then use "ttcp -t -P6 ::ffff:172.31.42.50" on YOWZA to send a stream of packets through the translator. Or vice-versa to stress IPv4 to IPv6 translation.
Known Issues and Limitations
- Can not send IPv4 fragments (as noted above).
- IP addresses embedded in an applications data stream (e.g., FTP) are not translated.
- DNS entries are not translated.
- No support for dynamic address assignment, beyond stateless address assignment.
- Bad things happen if you start Napt on a system without a functioning MSRIPv6 stack.
In addition to the regular contact information provided in the MSRIPv6 readme, feel free to contact the author directly with comments and questions regarding the implementation and configuration of the translator. You can reach the author by sending email to firstname.lastname@example.org.
- E. Nordmark. Stateless IP/ICMP Translator Algorithm (SIIT). IETF Request for Comments 2765, February 2000. Available as rfc2765.txt from your favorite IETF Internet Draft repository.
- G. Tsirtsis and P. Srishuresh. Network Address Translation - Protocol Translation (NAT-PT). IETF Request for Comments 2766, February 2000. Available as rfc2766.txt from your favorite IETF Internet Draft repository.
- M.E. Fiuczynski, V.K. Lam, B.N. Bershad. The Design and Implementation of an IPv6/IPv4 Network Address and Protocol Translator. In Proceedings of the 1998 Summer USENIX Conference, New Orleans, LA., June 1998. Also available here.