NAME

firehol-nat - set up NAT and port redirections

SYNOPSIS

{ nat to-destination | dnat [to] } ipaddr[:port] [random] [persistent] [id id] [at chain] [rule-params]

{ nat to-source | snat [to] } ipaddr[:port] [random] [persistent] [id id] [at chain] [rule-params]

{ nat redirect-to | redirect [to] } port[-range] [random] [id id] [at chain] [rule-params]

DESCRIPTION

Destination NAT is provided by nat to-destination and its synonym dnat.

Source NAT is provided by nat to-source and its synonym snat.

Redirection to a port on the local host is provided by nat redirect-to and its synonym redirect.

The port part of the new address is optional with SNAT and DNAT; if not specified it will not be changed.

When you apply NAT to a packet, the Linux kernel will track the changes it makes, so that when it sees replies the transformation will be applied in the opposite direction. For instance if you changed the destination port of a packet from 80 to 8080, when a reply comes back, its source is set as 80. This means the original sender is not aware a transformation is happening.

This means that NAT is only applied on the first packet of each connection (the nat FireHOL helper always appends state NEW to NAT statements).

The NAT helper can be used to setup load balancing. Check the section BALANCING below.

Note

The rule-params are used only to determine the traffic that will be matched for NAT in these commands, not to permit traffic to flow.

Applying NAT does not automatically create rules to allow the traffic to pass. You will still need to include client or server entries in an interface or router to allow the traffic.

When using dnat or redirect, the transformation is in the PREROUTING chain of the NAT table and happens before normal rules are matched, so your client or server rule should match the "modified" traffic.

When using snat, the transformation is in the POSTROUTING chain of the NAT table and happens after normal rules are matched, so your client or server rule should match the "unmodified" traffic.

See the netfilter flow diagram if you would like to see how network packets are processed by the kernel in detail.

The at keyword allows setting a different chain to attach the rules. For dnat and redirect the default is PREROUTING, but OUTPUT is also supported. For snat the default is POSTROUTING, but INPUT is also supported.

random will randomise the port mapping involved, to ensure the ports used are not predictable.

persistent is used when the statement is given alternatives (i.e. many destination servers for dnat, many source IPs for snat, many ports for redirect). It will attempt to keep each client on the same nat map. See below for more information about persistence.

The nat helper takes one of the following sub-commands:

to-destination ipaddr[:port]

Defines a Destination NAT (DNAT). Commonly thought of as port-forwarding (where packets destined for the firewall with a given port and protocol are sent to a different IP address and possibly port), DNAT is much more flexible in that any number of parameters can be matched before the destination information is rewritten.

ipaddr[:port] is the destination address to be set in packets matching rule-params.

If no rules are given, all forwarded traffic will be matched. outface should not be used in DNAT since the information is not available at the time the decision is made.

ipaddr[:port] accepts any --to-destination values that iptables(8) accepts. Run iptables -j DNAT --help for more information. Multiple ipaddr[:port] may be specified by separating with spaces and enclosing with quotes.

to-source ipaddr[:port]

Defines a Source NAT (SNAT). SNAT is similar to masquerading but is more efficient for static IP addresses. You can use it to give a public IP address to a host which does not have one behind the firewall. See also firehol-masquerade(5).

ipaddr[:port] is the source address to be set in packets matching rule-params.

If no rules are given, all forwarded traffic will be matched. inface should not be used in SNAT since the information is not available at the time the decision is made.

ipaddr[:port] accepts any --to-source values that iptables(8) accepts. Run iptables -j SNAT --help for more information. Multiple ipaddr[:port] may be specified by separating with spaces and enclosing with quotes.

redirect-to port[-range]

Redirect matching traffic to the local machine. This is typically useful if you want to intercept some traffic and process it on the local machine.

port[-range] is the port range (from-to) or single port that packets matching rule-params will be redirected to.

If no rules are given, all forwarded traffic will be matched. outface should not be used in REDIRECT since the information is not available at the time the decision is made.

BALANCING

NAT can balance multiple servers (or IPs in case of snat) when a range is specified. This is handled by the kernel.

Example:

dnat4 to 10.0.0.1-10.0.0.10 persistent proto tcp dst 1.1.1.1 dport 80

In the above example, the Linux kernel will give a persistent server to all the sockets of any single client.

FireHOL can also setup balancing using a round-robin or weighted average distribution of requests. However persistent cannot be used (the Linux kernel applies persistence on a single NAT statement).

Round Robin distribution

To enable round robin distribution, give multiple to values, space separated and enclosed in quotes, or comma separated.

Example:

dnat4 to 10.0.0.1,10.0.0.2,10.0.0.3 proto tcp dst 1.1.1.1 port 80
 # or
 dnat4 to "10.0.0.1 10.0.0.2 10.0.0.3" proto tcp dst 1.1.1.1 port 80

Ports can also be given per IP:

dnat4 to 10.0.0.1:70,10.0.0.2:80,10.0.0.3:90 proto tcp dst 1.1.1.1 port 80
 # or
 dnat4 to "10.0.0.1:70 10.0.0.2:80 10.0.0.3:90" proto tcp dst 1.1.1.1 port 80

Weighted distribution

To enable weighted distribution, append a slash with the weight requested for each entry.

FireHOL adds all the weights given and calculates the percentage of traffic each entry should receive.

Example:

dnat4 to 10.0.0.1/30,10.0.0.2/30,10.0.0.3/40 proto tcp dst 1.1.1.1 port 80
 # or
 dnat4 to "10.0.0.1/30 10.0.0.2/30 10.0.0.3/40" proto tcp dst 1.1.1.1 port 80
 # or
 dnat4 to 10.0.0.1:70/30,10.0.0.2:80/30,10.0.0.3:90/40 proto tcp dst 1.1.1.1 port 80
 # or
 dnat4 to "10.0.0.1:70/30 10.0.0.2:80/30 10.0.0.3:90/40" proto tcp dst 1.1.1.1 port 80

PERSISTENCE

The kernel supports persistence only if the NAT alternatives are contiguous (i.e. dnat to A-B, snat to A-B, redirect to 1000:1010, etc). If they are contiguous, persistence is left at the kernel. FireHOL does nothing.

If the alternatives are not contiguous, FireHOL will use the recent iptables module to apply persistence itself.

FireHOL supports mixed mode persistence. For example, you can have something like this:

dnat to A-B/70,C-D/20,F/10 persistence id mybalancer

The above is a weighted distribution of persistence. Group A-B will get 70%, C-D 20% and server F 10%.

Using the above, FireHOL will apply its persistence to pick one of the groups A-B, or C-D, or F. Once the group has been picked by FireHOL, the kernel will apply persistence within the group, to pick the server that will handle the request.

The FireHOL persistence works like this:

  1. A packet is received that should be NATed
  2. A lookup is made using the recent module to find if it has been seen before. The source IP of packet is looked up.
  3. If it has been seen before, the connection is mapped the same way the last time was mapped. The recent module is updated too.
  4. If it has not been seen before, the connection is mapped using the distribution method specified. The recent module is updated too, to be ready for the next connection.

The recent module has a few limitations:

  1. It has lookup tables. We need one lookup table for each member of of the NAT. FireHOL uses the id parameter and the definition of each alternative in the NAT statement to form a name for the lookup table. These lookup tables are persistent to firewall restarts, this is why FireHOL requires from you to set an id.

  2. It can keep entries in its lookup tables for a given time. FireHOL sets this to 3600 seconds. You can control it by setting FIREHOL_NAT_PERSISTENCE_SECONDS.

  3. It has a limit on the number of entries in the lookup tables. FireHOL cannot set this. This is kernel module option. The default is 200 entries.

Check this:

~~ # modinfo xt_recent filename: /lib/modules/4.1.12-gentoo/kernel/net/netfilter/xt_recent.ko alias: ip6t_recent alias: ipt_recent license: GPL description: Xtables: "recently-seen" host matching author: Jan Engelhardt

author: Patrick McHardy depends: x_tables intree: Y vermagic: 4.1.12-gentoo SMP preempt mod_unload modversions parm: ip_list_tot:number of IPs to remember per list (uint) parm: ip_list_hash_size:size of hash table used to look up IPs (uint) parm: ip_list_perms:permissions on /proc/net/xt_recent/* files (uint) parm: ip_list_uid:default owner of /proc/net/xt_recent/* files (uint) parm: ip_list_gid:default owning group of /proc/net/xt_recent/* files (uint) parm: ip_pkt_list_tot:number of packets per IP address to remember (max. 255) (uint) ~~
You have to consult your distribution documentation to set these.
You can find their current values by examining files found in
`/sys/module/xt_recent/parameters/` Unfortunately, these files
are not writable, so to change parameters you have unload and
reload the module (i.e. apply a firewall that does not use the
*recent* module, `rmmod xt_recent`, change the parameter,
re-apply a firewall that uses the *recent* module).

Normaly, you will need a line in `/etc/modprobe.d/netfitler.conf`
like this:

~~~~
options xt_recent ip_list_tot=16384
~~~~

The number 16384 I used is the max number of unique client IPs
I expect to have per hour (`FIREHOL_NAT_PERSISTENCE_SECONDS`)
for this service.

`ip_list_hash_size` is calculated by kernel when the module
is loaded to be bigger and up to twice `ip_list_tot`.

Once you have the balancer running, you can find its lookup tables in /proc/net/xt_recent/. There you will find files starting with the id parameter, one file for every alternative of the NAT rule.

EXAMPLES

# Port forwarding HTTP
 dnat4 to 192.0.2.2 proto tcp dport 80

 # Port forwarding HTTPS on to a different port internally
 dnat4 to 192.0.2.2:4443 proto tcp dport 443

 # Fix source for traffic leaving the firewall via eth0 with private address
 snat4 to 198.51.100.1 outface eth0 src 192.168.0.0/24

 # Transparent squid (running on the firewall) for some hosts
 redirect4 to 8080 inface eth0 src 198.51.100.0/24 proto tcp dport 80

 # Send to 192.0.2.1
 #  - all traffic arriving at or passing through the firewall
 nat4 to-destination 192.0.2.1

 # Send to 192.0.2.1
 #  - all traffic arriving at or passing through the firewall
 #  - which WAS going to 203.0.113.1
 nat4 to-destination 192.0.2.1 dst 203.0.113.1

 # Send to 192.0.2.1
 #  - TCP traffic arriving at or passing through the firewall
 #  - which WAS going to 203.0.113.1
 nat4 to-destination 192.0.2.1 proto tcp dst 203.0.113.1

 # Send to 192.0.2.1
 #  - TCP traffic arriving at or passing through the firewall
 #  - which WAS going to 203.0.113.1, port 25
 nat4 to-destination 192.0.2.1 proto tcp dport 25 dst 203.0.113.1

SEE ALSO