Tundra-NAT64 is an open-source IPv6-to-IPv4 & IPv4-to-IPv6 translator for Linux which operates entirely in user-space, can run in multiple threads (and thus make use of today's modern multicore CPUs) and uses either the TUN driver or inherited file descriptors to receive and send packets. It is written in pure C and translates packets according to the rules of SIIT (Stateless IP/ICMP Translation Algorithm; RFC 7915), while offering the following configurable address translation modes:
-
Stateless NAT64 – In this mode, Tundra is making it possible for a single host, or, in cooperation with Linux's in-kernel NAT66 translator (as described below), for any number of hosts on an IPv6-only network to access IPv4-only hosts.
-
Stateless CLAT – In this mode, Tundra is making it possible for programs using IPv4-only sockets (
AF_INET) to access IPv4-only hosts when running on a computer connected to an IPv6-only network with a NAT64 service. In addition, when running on a router which is connected to the outside world over an IPv6-only network with a NAT64 service, Tundra may be used to create a dual-stack internal network in cooperation with Linux's in-kernel NAT44 translator. -
SIIT – In this mode, Tundra is translating IPv6 packets whose addresses are composed of an IPv4 address wrapped inside a translation prefix into IPv4 packets with the same IPv4 addresses (extracted from the aforementioned prefix), and vice versa.
-
External – In this mode, Tundra delegates address translation to another program–server, with which it communicates via inherited file descriptors, Unix stream sockets or TCP. Tundra will translate packets from IPv4 to IPv6 and vice versa as per the rules of SIIT while querying an external address translator for IP addresses to be put in the translated packets. The specification of the protocol which Tundra uses to communicate with the external translator can be found in external_addr_xlat/EXTERNAL-ADDR-XLAT-PROTOCOL.md.
More information about the aforementioned address translation modes (including how to configure them) can be found in relevant sections of the example configuration file.
The SIIT/NAT64/CLAT translator offered by Tundra focuses on being minimal – features which are not necessary for SOHO-grade SIIT/NAT64/CLAT service are often omitted, and so are features which can be substituted by something else. One of the consequences of this design approach is that the program does not allocate any extra memory after it initializes.
Probably the most significant trait of this program, which makes it different from other NAT64/CLAT implementations, is that Tundra itself cannot act as a NAT64/CLAT translator for more than one host, as it lacks an internal dynamic address pool (or something similar) from which it would assign IP addresses to hosts needing its service – it uses a fixed single IP address specified in a configuration file instead (see the example config file for details). However, it can be used in cooperation with Linux's in-kernel NAT66/NAT44 translator and therefore translate traffic from any number of hosts/networks.
Tundra is, in certain aspects, similar to TAYGA (another stateless out-of-kernel NAT64
implementation, which can act as a CLAT translator in cooperation with clatd),
but there are some differences. Tundra is multi-threaded, has configurable address translation modes (including the
non-traditional external mode), can receive and send packets from inherited file descriptors, and lacks the
aforementioned dynamic address pool. TAYGA also inspired this program's name - both "taiga" and "tundra" are subarctic
biomes and the word "tundra" starts with "tun", the name of the driver which Tundra uses to exchange packets.
The reason why this program is named Tundra-NAT64 despite it offering other address translation modes than just NAT64
is that originally, it was meant to be only a stateless NAT64 translator and the other modes were added later. Since
rebranding it would be quite difficult and could cause certain problems (due to, for example, permalinks pointing to
this Git repository whose name/URL contains the nat64), I decided not to do it.
Tundra was the subject of a Czech-language talk I held at the Seminář IPv6: deset let poté conference on the 6th June 2022 in Prague – presentation, video. In addition, if you happen to be interested, you may have a look at this post on my English-language blog, in which the changes to this program made between the versions 4.1.6 and 5.0.1 are summed up.
Since Tundra has no dependencies other than Linux's standard C library and libpthread, it can be compiled by a single
command and without the use of a build system:
gcc -Wall -Wextra -pthread -std=c11 -O3 -flto -o tundra-nat64 src/t64_*.cBoth gcc and clang may be used to compile the program.
Tundra loads its settings from a configuration file. This repository contains a sample configuration file with detailed comments: tundra-nat64.example.conf.
The output of ./tundra-nat64 --help is as follows:
Usage: ./tundra-nat64 [OPTION]... [MODE_OF_OPERATION]
Options:
-h, --help
Prints help and exits.
-v, --version
Prints version information and exits.
-l, --license
Prints license and exits.
-c, --config-file=CONFIG_FILE_PATH
Specifies the file from which the program's configuration will be loaded.
DEFAULT: /etc/tundra-nat64/tundra-nat64.conf
NOTE: To load the configuration from standard input, specify '-' as the config file path.
-f, --io-inherited-fds=THREAD1_IN,THREAD1_OUT[;THREAD2_IN,THREAD2_OUT]...
Specifies the file descriptors to be used in the 'inherited-fds' I/O mode. Ignored otherwise.
-F, --addressing-external-inherited-fds=THREAD1_IN,THREAD1_OUT[;THREAD2_IN,THREAD2_OUT]...
Specifies the file descriptors to be used for the 'inherited-fds' transport of the 'external' addressing mode. Ignored otherwise.
Modes of operation:
translate
The program will act as a stateless NAT64/CLAT translator.
This is the default mode of operation.
mktun
Creates a persistent TUN device according to the configuration file, then exits.
Applicable only in the 'tun' I/O mode.
rmtun
Destroys a previously created persistent TUN device according to the configuration file, then exits.
Applicable only in the 'tun' I/O mode.
validate-config
Tries to configure the program and prints an informational message if it succeeds, then exits.
print-config
Prints the program's configuration in a human-readable format, then exits.
The following example shows how Tundra could be deployed as NAT64 translator for an IPv6-only network with the use of the example configuration file on a router which has access to both IPv4 and IPv6:
TUNDRA_CONFIG_FILE="./tundra-nat64.example.conf"
WAN_INTERFACE_NAME="eth0" # Remember to adjust this!
# Create a new TUN network interface
./tundra-nat64 --config-file=$TUNDRA_CONFIG_FILE mktun
# Set up the TUN interface
ip link set dev tundra up
ip addr add 192.168.64.254/24 dev tundra # The IP address should be different from 'translator.ipv4' and 'router.ipv4'!
ip -6 addr add fd00:6464::fffe/64 dev tundra # The IP address should be different from 'translator.ipv6' and 'router.ipv6'!
ip -6 route add 64:ff9b::/96 dev tundra
ip6tables -t nat -A POSTROUTING -d 64:ff9b::/96 -o tundra -j SNAT --to-source=fd00:6464::2 # On some kernels, the support for NAT66 may need to be installed as a module
iptables -t nat -A POSTROUTING -o $WAN_INTERFACE_NAME -j MASQUERADE # Perform NAT44 on all packets going to the internet, including the ones generated by Tundra
# Start the NAT64 translator (to terminate it, send SIGTERM, SIGINT or SIGHUP to the process)
./tundra-nat64 --config-file=$TUNDRA_CONFIG_FILE translate
# After the translator has terminated, remove the previously added NAT rules...
ip6tables -t nat -D POSTROUTING -d 64:ff9b::/96 -o tundra -j SNAT --to-source=fd00:6464::2
iptables -t nat -D POSTROUTING -o $WAN_INTERFACE_NAME -j MASQUERADE
# ... and remove the TUN interface
./tundra-nat64 --config-file=$TUNDRA_CONFIG_FILE rmtunTo make hosts on IPv6-only networks use the NAT64 translator to access IPv4-only services, you will need to provide them
with a DNS64 recursive resolver. You can either deploy your own one (all major recursive DNS servers, such as
Unbound or
Knot Resolver, nowadays support DNS64 natively),
or, if your NAT64 deployment uses the well-known prefix of 64:ff9b::/96, use a public DNS64 resolver, such as the
one provided by Google or
Cloudflare.
This repository contains the following configuration examples for specific platforms and translation modes:
This project is licensed under the 3-clause BSD license – see the LICENSE file.
Programmed by Vít Labuda.