Skip to the content.


This chapter will describe NPF configuration in the context of NetBSD or a similar UNIX-like system environment.

The standalone NPF configuration would be constructed using the libnpf library and submitted to the kernel-side component, using the npfkern library. See the libnpf(3) and npfkern(3) manual pages for the API.

NetBSD beginners can consult rc(8), rc.conf(5), ifconfig.if(5), route(8) and other manual pages for the general networking configuration in the system.

Structure and syntax

The NPF configuration is represented by a file, called npf.conf (the default location is /etc/npf.conf). It is loaded and NPF is generally operated via the command line utility called npfctl. For a reference, use npf.conf(5) and npfctl(8) manual pages.

There are multiple structural elements that npf.conf may contain, such as:


Variables are specified using the dollar ($) sign, which is used for both definition and referencing of a variable. Variables are defined by assigning a value to them as follows:

$var1 =

A variable may also be defined as a set:

$var2 = {, }

Common variable definitions are for IP addresses, networks, ports, and interfaces.


Tables are specified using a name between angle brackets < and >. The following is an example of table definition:

table <blocklist> type ipset

Currently, tables support three data storage types: ipset, lpm, or const. The contents of the table may be pre-loaded from the specified file. The const tables are immutable (no insertions or deletions after loading) and therefore must always be loaded from a file.

The specified file should contain a list of IP addresses and/or networks in the form of or

Tables of type ipset and const can only contain IP addresses. The lpm tables can contain networks and they will perform the longest prefix match on lookup.


In NPF, an interface can be referenced directly by using its name, or can be passed to an extraction function which will return a list of IP addresses configured on the actual associated interface.

It is legal to pass an extracted list from an interface in keywords where NPF would expect instead a direct reference to said interface. In this case, NPF infers a direct reference to the interface, and does not con‐ sider the list.

There are two types of IP address lists. With a static list, NPF will capture the interface addresses on configuration load, whereas with a dynamic list NPF will capture the runtime list of addresses, reflecting any changes to the interface, including the attach and detach. Note that with a dynamic list, bringing the interface down has no effect, all addresses will remain present.

The following functions exist, to extract addresses from an interface with a chosen list type and IP address type:

Function Type Description
inet4(interface) static list IPv4 addresses
inet6(interface) static list IPv6 addresses
ifaddrs(interface) dynamic list Both IPv4 and IPv6. The family keyword of a filtering rule can be used in combination to explicitly select an IP address type.

Example of configuration:

$var1 = inet4(wm0)
$var2 = ifaddrs(wm0)

group default {
  block in on wm0 all               # rule 1
  block in on $var1 all             # rule 2
  block in on inet4(wm0) all        # rule 3
  pass in on inet6(wm0) from $var2  # rule 4
  pass in on wm0 from ifaddrs(wm0)  # rule 5

In the above example, $var1 is the static list of IPv4 addresses configured on wm0, and $var2 is the dynamic list of all the IPv4 and IPv6 addresses configured on wm0. The first three rules are equivalent, because with the block ... on <interface> syntax, NPF expects a direct reference to an interface, and therefore does not consider the extraction functions. The fourth and fifth rules are equivalent, for the same reason.


NPF requires that all rules be defined within groups. Groups can be thought of as higher level rules which can contain subrules. Groups may have the following options: name, interface, and direction. Packets matching group criteria are passed to the ruleset of that group. If a packet does not match any group, it is passed to the default group. The default group must always be defined.

Example of configuration:

group "my‐name" in on wm0 {
  # List of rules, for packets received on wm0

group default {
  # List of rules, for the other packets


With a rule statement NPF is instructed to pass or block a packet depending on packet header information, transit direction and the interface it arrived on, either immediately upon match or using the last match.

If a packet matches a rule which has the final option set, this rule is considered the last matching rule, and evaluation of subsequent rules is skipped. Otherwise, the last matching rule is used.

The proto keyword can be used to filter packets by layer 4 protocol (TCP, UDP, ICMP or other). Its parameter should be a protocol number or its symbolic name, as specified in the /etc/protocols file. This keyword can additionally have protocol‐specific options, such as flags.

The flags keyword can be used to match the packets against specific TCP flags, according to the following syntax:

proto tcp flags match[/mask]

Where match is the set of TCP flags to be matched, out of the mask set, both sets being represented as a string combination of: S (SYN), A (ACK), F (FIN), and R (RST). The flags that are not present in mask are ignored.

To notify the sender of a blocking decision, three return options can be used in conjunction with a block rule:

Keyword Description
return Behaves as return‐rst or return‐icmp, depending on whether the packet being blocked is TCP or UDP.
return‐rst Return a TCP RST message, when the packet being blocked is a TCP packet. Applies to IPv4 and IPv6.
return‐icmp Return an ICMP UNREACHABLE message, when the packet being blocked is a UDP packet. Applies to IPv4 and IPv6.

Further packet specification at present is limited to TCP and UDP under‐ standing source and destination ports, and ICMP and IPv6‐ICMP understand‐ ing icmp‐type.

A rule can also instruct NPF to create an entry in the state table when passing the packet or to apply a procedure to the packet (e.g. “log”).

A “fully‐featured” rule would for example be:

pass stateful in final family inet4 proto tcp flags S/SA \
  from $source port $sport to $dest port $dport apply "someproc"

Alternatively, NPF supports pcap-filter(7) syntax, for example:

block out final pcap‐filter "tcp and dst"

Fragments are not selectable since NPF always reassembles packets before further processing.


Stateful packet inspection is enabled using the stateful or stateful‐ends keywords. The former creates a state which is uniquely identified by a 5‐tuple (source and destination IP addresses, port numbers and an inter‐ face identifier). The latter excludes the interface identifier and must be used with precaution. In both cases, a full TCP state tracking is performed for TCP connections and a limited tracking for message‐based protocols (UDP and ICMP).

By default, a stateful rule implies SYN‐only flag check (“flags S/SAFR”) for the TCP packets. It is not advisable to change this behavior; how‐ ever, it can be overridden with the aforementioned flags keyword.


Network Address Translation (NAT) is expressed in a form of segment map‐ ping. The translation may be dynamic (stateful) or static (stateless). The following mapping types are available:

Syntax Description
‐> outbound NAT (translation of the source)
<‐ inbound NAT (translation of the destination)
<‐> bi‐directional NAT (combination of inbound and outbound NAT)

The following would translate the source ( to the IP address specified by $pub_ip for the packets on the interface $ext_if.

map $ext_if dynamic ‐> $pub_ip

Translations are implicitly filtered by limiting the operation to the network segments specified, that is, translation would be performed only on packets originating from the network. Explicit filter criteria can be specified using pass criteria ... as an additional option of the mapping.

The dynamic NAT implies network address and port translation (NAPT). The port translation can be controlled explicitly. For example, the following provides “port forwarding”, redirecting the public port 9022 to the port 22 of an internal host:

map $ext_if dynamic proto tcp port 22 <‐ $ext_if port 9022

If the dynamic NAT is configured with multiple translation addresses, then a custom selection algorithm can be chosen using the algo keyword. The currently available algorithms are:

Algorithm Description
ip-hash The translation address for a new connection is selected based on a hash of the original source and destination addresses. This algorithms attempts to keep all connections of particular client associated with the same translation address. This is the default algorithm.
round-robin The translation address for each new connection is selected on a round-robin basis.

The static NAT can also have different address translation algorithms, chosen using the algo keyword. The currently available algorithms are:

Algorithm Description
netmap Network address mapping from one segment to another, leaving the host part as-is. The new address is computed as following: addr = net-addr | (orig-addr & ~mask)
npt66 IPv6‐to‐IPv6 network prefix translation (NPTv6)

If no algorithm is specified, then 1:1 address mapping is assumed. Currently, the static NAT algorithms do not perform port translation.

Application Level Gateways

Certain application layer protocols are not compatible with NAT and require translation outside layers 3 and 4. Such translation is per‐ formed by packet filter extensions called Application Level Gateways (ALGs).

NPF supports the following ALGs:

The ALGs are built‐in. If NPF is used as kernel module, then they come as kernel modules too. In such case, the ALG kernel modules can be autoloaded through the configuration, using the alg keyword.

For example:

alg "icmp"

Alternatively, the ALG kernel modules can be loaded manually, using modload(8).


A rule procedure is defined as a collection of extension calls (it may have none). Every extension call has a name and a list of options in the form of key‐value pairs. Depending on the call, the key might represent the argument and the value might be optional. Available options:

The available normalization options are:

Parameter Description
max‐mss <value> Enforce a maximum value for the Maximum Segment Size (MSS) TCP option. Typically, for “MSS clamping”.
min‐ttl <value> Enforce a minimum value for the IPv4 Time To Live (TTL) parameter.
no‐df Remove the Don’t Fragment (DF) flag from IPv4 packets.
random‐id Randomize the IPv4 ID parameter.

For example:

procedure "someproc" {
  log: npflog0
  normalize: "random‐id", "min‐ttl" 64, "max‐mss" 1432

In this case, the procedure calls the logging and normalization modules.


Text after a hash (#) character is considered a comment. The backslash (\) character at the end of a line marks a continuation line, i.e., the next line is considered an extension of the present line.

Control and operation

NPF is controlled using the npfctl(8) utility.

$ npfctl
Usage:	npfctl start | stop | flush | show | stats
	npfctl validate | reload [<rule-file>]
	npfctl rule "rule-name" { add | rem } <rule-syntax>
	npfctl rule "rule-name" rem-id <rule-id>
	npfctl rule "rule-name" { list | flush }
	npfctl table "table-name" { add | rem | test } <address/mask>
	npfctl table "table-name" { list | flush }
	npfctl table "table-name" replace [-n "name"] [-t <type>] <table-file>
	npfctl save | load
	npfctl list [-46hNnw] [-i <ifname>]
	npfctl debug [<rule-file>] [<raw-output>]

Once the NPF configuration file has been written, use npfctl to load it and then start the packet handling:

$ npfctl load
$ npfctl start

Any modifications of npf.conf require reloading of the ruleset by performing a reload command in order to make the changes active. One difference from other packet filters is the behaviour of the start and stop commands. These commands do not actually change (i.e. load or unload) the active configuration. Running start will only enable the passing of packets through NPF, while stop will disable such passing. Therefore, configuration should first be activated using the reload command and then filtering enabled with start. Similarly, clearing of the active configuration is done by performing the stop and flush commands. Such behaviour allows users to efficiently disable and enable filtering without actually changing the active configuration, as it may be unnecessary.

Autostart on boot

In NetBSD, the rc.d system can be used to start NPF on boot. The following is an example for starting NPF and loading the configuration through the rc.d script:

$ echo 'npf=YES' >> /etc/rc.conf
$ /etc/rc.d/npf reload
Reloading NPF ruleset.
$ /etc/rc.d/npf start
Enabling NPF.