Variables

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:

A variable may also be defined as a set:

$var2 = {
    10.0.0.1,   # First host
    10.0.0.2,   # Second host
}

Newlines within curly braces are ignored, and trailing commas are optional. Common variable definitions are for IP addresses, networks, ports, and interfaces.

Tables

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

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 10.1.1.1 or 10.0.0.0/24.

Tables of type ipset and const can only contain IP addresses (without masks). The lpm tables can contain networks and they will perform the longest prefix match on lookup.

Interfaces

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 consider 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.

Three functions exist, to extract addresses from an interface with a chosen list type and 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.

Groups

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
}

Rules

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:

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:

The from and to keywords are provided to filter by source or destination IP addresses. They can be used in conjunction with the port keyword. Negation (the exclamation mark) can be used in front of the address filter criteria.

Further packet specification at present is limited to TCP and UDP understanding source and destination ports, and ICMP and IPv6-ICMP understanding 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:

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

Layer 2 filtering

NPF allows the filtering of frames at the data link layer. NPF also requires that the inspection rules are defined within groups. Groups containing rules to filter frames should be marked with a layer-2 label. If layer 2 groups are defined in your NPF configuration, then a layer-2 default group becomes mandatory.

Example of configuration:

group "my-name" in on wm0 layer-2 {
	# List of rules, for frames received on wm0
}
group default layer-2 {
	# List of rules, for the other frames
}

Rules for filtering at the data link layer are configured based on the interface name, direction, source and destination MAC addresses, and EtherType. Rules that are defined for the link layer should pass the ether keyword after the pass or block instruction. EtherType is passed on the rule by preceeding the four digit hexadecimal constant EtherType with "Ex". When a frame matches a rule with the final keyword, the rule is considered the last matching rule.

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

pass ether in final from $src_mac to $dest_mac type $ether_type

Example of rule configuration:

pass ether from 00:00:5E:00:53:00 to 00:00:5E:00:53:01 type Ex0800

This passes frames with source MAC address 00:00:5E:00:53:00 and destination MAC address 00:00:5E:00:53:01 carrying IP packets(0800). Alternatively, layer 2 rules also support variables.

Filtering at this layer is stateless and has no access to upper layer protocols. Block returns are not supported. MAC address tables are also not supported yet.

User/group ID filtering

NPF allows filtering by user or group identity. Packet filtering by user or group controls data packet flows based on the user or group identity of the process that generated the traffic, or is waiting to receive traffic, rather than just traditional parameters like IP address, port number, and protocol.

There are many situations where this is useful:

• One can allow specific destinations to be accessed only by certain users or groups.
• Two processes are using a specific port, but only one should be allowed to access packets originating from a particular host.
• Prevent an untrusted user from making any network connections.
• A user application that has been exploited can be prevented from making network connections to a command-and-control server.
• Can restrict access to particular networks to network administrators only.

The above rule only allows sockets of processes owned by user jack and belonging to a group with an ID value of less than 1000.

The above rule prevents all listening sockets bound by processes owned by any user with the ID value greater than 100 and belonging to the wheel group.

A rule can have either a user ID or group ID set. If both are set, both must agree to be a match to the socket involved in communication. Numbers or names can be used for the identification of the user or group as they still resolve to a numeric ID of the user or group.

Stateful

NPF supports stateful packet inspection which can be used to bypass unnecessary rule processing as well as to complement NAT. The connection state is uniquely identified by an n-tuple: IP version, layer 4 protocol, source and destination IP addresses and port numbers. Each state is represented by two keys: one for the original flow and one for the reverse flow, so that the reverse lookup on the returning packets would succeed. The packets are matched against the connection direction respectively.

Depending on the settings (see the section on state.key in the npf-params(7) manual), the connection identifier (keys) may also include the interface ID, making the states per-interface.

Stateful packet inspection is enabled using the stateful or stateful-all keywords. The former matches the interface after the state lookup, while the latter avoids matching the interface (assuming the state.key.interface parameter is disabled), i.e. making the state global, and must be used with caution. 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; however, it can be overridden with the aforementioned flags keyword.

Map

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

The following would translate the source (10.1.1.0/24) to the IP address specified by $pub_ip for the packets on the interface $ext_if.

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 10.1.1.0/24 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:

In the regular dynamic NAT case, it is also possible to disable port translation using the no-ports flag.

The translation address can also be dynamic, based on the interface. The following would select the IPv4 address(es) currently assigned to the interface:

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 for the dynamic translation are:

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

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 performed 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:

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

Procedures

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:

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.

Parameter settings

NPF supports a set of dynamically tunable configuration-wide parameters. For example:

set state.tcp.timeout.time_wait 0  # destroy the state immediately

See npf-params(7) for the list of parameters and their details.

Misc

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. Additionally, within curly braces of variable definitions, newlines are allowed without continuation characters.