[OpenBSD]

[Previous: Tables] [Contents] [Next: Network Address Translation]

PF: Packet Filtering

Table of Contents

Introduction

Packet filtering is the selective passing or blocking of data packets as they pass through a network interface. The criteria that pf(4) uses when inspecting packets are based on the Layer 3 (IPv4 and IPv6) and Layer 4 (TCP, UDP, ICMP, and ICMPv6) headers. The most often used criteria are source and destination address, source and destination port, and protocol.

Filter rules specify the criteria that a packet must match and the resulting action, either block or pass, that is taken when a match is found. Filter rules are evaluated in sequential order, first to last. Unless the packet matches a rule containing the quick keyword, the packet will be evaluated against all filter rules before the final action is taken. The last rule to match is the "winner" and will dictate what action to take on the packet. There is an implicit pass all at the beginning of a filtering ruleset meaning that if a packet does not match any filter rule the resulting action will be pass.

Rule Syntax

The general, highly simplified syntax for filter rules is:
action [direction] [log] [quick] [on interface] [af] [proto protocol] \
   [from src_addr [port src_port]] [to dst_addr [port dst_port]] \
   [flags tcp_flags] [state]
action
The action to be taken for matching packets, either pass or block. The pass action will pass the packet back to the kernel for further processing while the block action will react based on the setting of the block-policy option. The default reaction may be overridden by specifying either block drop or block return.
direction
The direction the packet is moving on an interface, either in or out.
log
Specifies that the packet should be logged via pflogd(8). If the rule creates state then only the packet which establishes the state is logged. To log all packets regardless, use log (all).
quick
If a packet matches a rule specifying quick, then that rule is considered the last matching rule and the specified action is taken.
interface
The name or group of the network interface that the packet is moving through. Interfaces can be added to arbitrary groups using the ifconfig(8) command. Several groups are also automatically created by the kernel: This would cause the rule to match for any packet traversing any ppp or carp interface, respectively.
af
The address family of the packet, either inet for IPv4 or inet6 for IPv6. PF is usually able to determine this parameter based on the source and/or destination address(es).
protocol
The Layer 4 protocol of the packet:
src_addr, dst_addr
The source/destination address in the IP header. Addresses can be specified as:
src_port, dst_port
The source/destination port in the Layer 4 packet header. Ports can be specified as:
tcp_flags
Specifies the flags that must be set in the TCP header when using proto tcp. Flags are specified as flags check/mask. For example: flags S/SA - this instructs PF to only look at the S and A (SYN and ACK) flags and to match if only the SYN flag is "on" (and is applied to all TCP rules by default). flags any instructs PF not to check flags.
state
Specifies whether state information is kept on packets matching this rule.

Default Deny

The recommended practice when setting up a firewall is to take a "default deny" approach. That is, to deny everything and then selectively allow certain traffic through the firewall. This approach is recommended because it errs on the side of caution and also makes writing a ruleset easier.

To create a default deny filter policy, the first two filter rules should be:

block in  all
block out all

This will block all traffic on all interfaces in either direction from anywhere to anywhere.

Passing Traffic

Traffic must now be explicitly passed through the firewall or it will be dropped by the default deny policy. This is where packet criteria such as source/destination port, source/destination address, and protocol come into play. Whenever traffic is permitted to pass through the firewall the rule(s) should be written to be as restrictive as possible. This is to ensure that the intended traffic, and only the intended traffic, is permitted to pass.

Some examples:

# Pass traffic in on dc0 from the local network, 192.168.0.0/24,
# to the OpenBSD machine's IP address 192.168.0.1. Also, pass the
# return traffic out on dc0.
pass in  on dc0 from 192.168.0.0/24 to 192.168.0.1
pass out on dc0 from 192.168.0.1 to 192.168.0.0/24


# Pass TCP traffic in on fxp0 to the web server running on the
# OpenBSD machine. The interface name, fxp0, is used as the
# destination address so that packets will only match this rule if
# they're destined for the OpenBSD machine.
pass in on fxp0 proto tcp from any to fxp0 port www

The quick Keyword

As indicated earlier, each packet is evaluated against the filter ruleset from top to bottom. By default, the packet is marked for passage, which can be changed by any rule, and could be changed back and forth several times before the end of the filter rules. The last matching rule "wins". There is an exception to this: The quick option on a filtering rule has the effect of canceling any further rule processing and causes the specified action to be taken. Let's look at a couple examples:

Wrong:

block in on fxp0 proto tcp to port ssh
pass  in all

In this case, the block line may be evaluated, but will never have any effect, as it is then followed by a line which will pass everything.

Better:

block in quick on fxp0 proto tcp to port ssh
pass  in all

These rules are evaluated a little differently. If the block line is matched, due to the quick option, the packet will be blocked, and the rest of the ruleset will be ignored.

Keeping State

One of Packet Filter's important abilities is "keeping state" or "stateful inspection". Stateful inspection refers to PF's ability to track the state, or progress, of a network connection. By storing information about each connection in a state table, PF is able to quickly determine if a packet passing through the firewall belongs to an already established connection. If it does, it is passed through the firewall without going through ruleset evaluation.

Keeping state has many advantages including simpler rulesets and better packet filtering performance. PF is able to match packets moving in either direction to state table entries meaning that filter rules which pass returning traffic don't need to be written. And, since packets matching stateful connections don't go through ruleset evaluation, the time PF spends processing those packets can be greatly lessened.

When a rule creates state, the first packet matching the rule creates a "state" between the sender and receiver. Now, not only do packets going from the sender to receiver match the state entry and bypass ruleset evaluation, but so do the reply packets from receiver to sender.

All pass rules automatically create a state entry when a packet matches the rule. This can be explicitly disabled by using the no state option.

pass out on fxp0 proto tcp from any to any

This rule allows any outbound TCP traffic on the fxp0 interface and also permits the reply traffic to pass back through the firewall. Keeping state significantly improves the performance of your firewall as state lookups are dramatically faster than running a packet through the filter rules.

The modulate state option works just like keep state except that it only applies to TCP packets. With modulate state, the Initial Sequence Number (ISN) of outgoing connections is randomized. This is useful for protecting connections initiated by certain operating systems that do a poor job of choosing ISNs. To allow simpler rulesets, the modulate state option can be used in rules that specify protocols other than TCP; in those cases, it is treated as keep state.

Keep state on outgoing TCP, UDP, and ICMP packets and modulate TCP ISNs:

pass out on fxp0 proto { tcp, udp, icmp } from any \
    to any modulate state

Another advantage of keeping state is that corresponding ICMP traffic will be passed through the firewall. For example, if a TCP connection passing through the firewall is being tracked statefully and an ICMP source-quench message referring to this TCP connection arrives, it will be matched to the appropriate state entry and passed through the firewall.

The scope of a state entry is controlled globally by the state-policy runtime option and on a per rule basis by the if-bound and floating state option keywords. These per rule keywords have the same meaning as when used with the state-policy option. Example:

pass out on fxp0 proto { tcp, udp, icmp } from any \
    to any modulate state (if-bound)

This rule would dictate that in order for packets to match the state entry, they must be transiting the fxp0 interface.

Keeping State for UDP

One will sometimes hear it said that, "One can not create state with UDP as UDP is a stateless protocol!" While it is true that a UDP communication session does not have any concept of state (an explicit start and stop of communications), this does not have any impact on PF's ability to create state for a UDP session. In the case of protocols without "start" and "end" packets, PF simply keeps track of how long it has been since a matching packet has gone through. If the timeout is reached, the state is cleared. The timeout values can be set in the options section of the pf.conf file.

Stateful Tracking Options

Filter rules that create state entries can specify various options to control the behavior of the resulting state entry. The following options are available:
max number
Limit the maximum number of state entries the rule can create to number. If the maximum is reached, packets that would normally create state fail to match this rule until the number of existing states decreases below the limit.
no state
Prevents the rule from automatically creating a state entry.
source-track
This option enables the tracking of number of states created per source IP address. This option has two formats: The total number of source IP addresses tracked globally can be controlled via the src-nodes runtime option.
max-src-nodes number
When the source-track option is used, max-src-nodes will limit the number of source IP addresses that can simultaneously create state. This option can only be used with source-track rule.
max-src-states number
When the source-track option is used, max-src-states will limit the number of simultaneous state entries that can be created per source IP address. The scope of this limit (i.e., states created by this rule only or states created by all rules that use source-track) is dependent on the source-track option specified.

Options are specified inside parenthesis and immediately after one of the state keywords (keep state, modulate state, or synproxy state). Multiple options are separated by commas. In OpenBSD 4.1 and later, the keep state option became the implicit default for all filter rules. Despite this, when specifying stateful options, one of the state keywords must still be used in front of the options.

An example rule:

pass in on $ext_if proto tcp to $web_server \
    port www keep state \
    (max 200, source-track rule, max-src-nodes 100, max-src-states 3)

The rule above defines the following behavior:

A separate set of restrictions can be placed on stateful TCP connections that have completed the 3-way handshake.

max-src-conn number
Limit the maximum number of simultaneous TCP connections which have completed the 3-way handshake that a single host can make.
max-src-conn-rate number / interval
Limit the rate of new connections to a certain amount per time interval.

Both of these options automatically invoke the source-track rule option and are incompatible with source-track global.

Since these limits are only being placed on TCP connections that have completed the 3-way handshake, more aggressive actions can be taken on offending IP addresses.

overload <table>
Put an offending host's IP address into the named table.
flush [global]
Kill any other states that match this rule and that were created by this source IP. When global is specified, kill all states matching this source IP, regardless of which rule created the state.

An example:

table <abusive_hosts> persist
block in quick from <abusive_hosts>

pass in on $ext_if proto tcp to $web_server \
    port www flags S/SA keep state \
    (max-src-conn 100, max-src-conn-rate 15/5, overload <abusive_hosts> flush)

This does the following:

TCP Flags

Matching TCP packets based on flags is most often used to filter TCP packets that are attempting to open a new connection. The TCP flags and their meanings are listed here:

To have PF inspect the TCP flags during evaluation of a rule, the flags keyword is used with the following syntax:

flags check/mask
flags any

The mask part tells PF to only inspect the specified flags and the check part specifies which flag(s) must be "on" in the header for a match to occur. Using the any keyword allows any combination of flags to be set in the header.

pass in on fxp0 proto tcp from any to any port ssh flags S/SA
pass in on fxp0 proto tcp from any to any port ssh

As flags S/SA is set by default, the above rules are equivalent, Each of these rules passes TCP traffic with the SYN flag set while only looking at the SYN and ACK flags. A packet with the SYN and ECE flags would match the above rules, while a packet with SYN and ACK or just ACK would not.

The default flags can be overridden by using the flags option as outlined above.

One should be careful with using flags -- understand what you are doing and why, and be careful with the advice people give as a lot of it is bad. Some people have suggested creating state "only if the SYN flag is set and no others". Such a rule would end with: 

     . . . flags S/FSRPAUEW  bad idea!!

The theory is, create state only on the start of the TCP session, and the session should start with a SYN flag, and no others. The problem is some sites are starting to use the ECN flag and any site using ECN that tries to connect to you would be rejected by such a rule. A much better guideline is to not specify any flags at all and let PF apply the default flags to your rules. If you truly need to specify flags yourself then this combination should be safe:

. . . flags S/SAFR

While this is practical and safe, it is also unnecessary to check the FIN and RST flags if traffic is also being scrubbed. The scrubbing process will cause PF to drop any incoming packets with illegal TCP flag combinations (such as SYN and RST) and to normalize potentially ambiguous combinations (such as SYN and FIN).

TCP SYN Proxy

Normally when a client initiates a TCP connection to a server, PF will pass the handshake packets between the two endpoints as they arrive. PF has the ability, however, to proxy the handshake. With the handshake proxied, PF itself will complete the handshake with the client, initiate a handshake with the server, and then pass packets between the two. The benefit of this process is that no packets are sent to the server before the client completes the handshake. This eliminates the threat of spoofed TCP SYN floods affecting the server because a spoofed client connection will be unable to complete the handshake.

The TCP SYN proxy is enabled using the synproxy state keywords in filter rules. Example:

pass in on $ext_if proto tcp to $web_server port www synproxy state

Here, connections to the web server will be TCP proxied by PF.

Because of the way synproxy state works, it also includes the same functionality as keep state and modulate state.

The SYN proxy will not work if PF is running on a bridge(4).

Blocking Spoofed Packets

Address "spoofing" is when a malicious user fakes the source IP address in packets they transmit in order to either hide their real address or to impersonate another node on the network. Once the user has spoofed their address they can launch a network attack without revealing the true source of the attack or attempt to gain access to network services that are restricted to certain IP addresses.

PF offers some protection against address spoofing through the antispoof keyword:

antispoof [log] [quick] for interface [af]
log
Specifies that matching packets should be logged via pflogd(8).
quick
If a packet matches this rule then it will be considered the "winning" rule and ruleset evaluation will stop.
interface
The network interface to activate spoofing protection on. This can also be a list of interfaces.
af
The address family to activate spoofing protection for, either inet for IPv4 or inet6 for IPv6.

Example:

antispoof for fxp0 inet

When a ruleset is loaded, any occurrences of the antispoof keyword are expanded into two filter rules. Assuming that interface fxp0 has IP address 10.0.0.1 and a subnet mask of 255.255.255.0 (i.e., a /24), the above antispoof rule would expand to:

block in on ! fxp0 inet from 10.0.0.0/24 to any
block in inet from 10.0.0.1 to any

These rules accomplish two things:

NOTE: The filter rules that the antispoof rule expands to will also block packets sent over the loopback interface to local addresses. It's best practice to skip filtering on loopback interfaces anyways, but this becomes a necessity when using antispoof rules:

set skip on lo0

antispoof for fxp0 inet

Usage of antispoof should be restricted to interfaces that have been assigned an IP address. Using antispoof on an interface without an IP address will result in filter rules such as:

block drop in on ! fxp0 inet all
block drop in inet all

With these rules there is a risk of blocking all inbound traffic on all interfaces.

Unicast Reverse Path Forwarding

PF offers a Unicast Reverse Path Forwarding (uRPF) feature. When a packet is run through the uRPF check, the source IP address of the packet is looked up in the routing table. If the outbound interface found in the routing table entry is the same as the interface that the packet just came in on, then the uRPF check passes. If the interfaces don't match, then it's possible the packet has had its source address spoofed.

The uRPF check can be performed on packets by using the urpf-failed keyword in filter rules:

block in quick from urpf-failed label uRPF

Note that the uRPF check only makes sense in an environment where routing is symmetric.

uRPF provides the same functionality as antispoof rules.

Passive Operating System Fingerprinting

Passive OS Fingerprinting (OSFP) is a method for passively detecting the operating system of a remote host based on certain characteristics within that host's TCP SYN packets. This information can then be used as criteria within filter rules.

PF determines the remote operating system by comparing characteristics of a TCP SYN packet against the fingerprints file, which by default is /etc/pf.os. Once PF is enabled, the current fingerprint list can be viewed with this command:

# pfctl -s osfp

Within a filter rule, a fingerprint may be specified by OS class, version, or subtype/patch level. Each of these items is listed in the output of the pfctl command shown above. To specify a fingerprint in a filter rule, the os keyword is used:

pass  in on $ext_if from any os OpenBSD keep state
block in on $ext_if from any os "Windows 2000"
block in on $ext_if from any os "Linux 2.4 ts"
block in on $ext_if from any os unknown

The special operating system class unknown allows for matching packets when the OS fingerprint is not known.

TAKE NOTE of the following:

IP Options

By default, PF blocks packets with IP options set. This can make the job more difficult for "OS fingerprinting" utilities like nmap. If you have an application that requires the passing of these packets, such as multicast or IGMP, you can use the allow-opts directive:
pass in quick on fxp0 all allow-opts

Filtering Ruleset Example

Below is an example of a filtering ruleset. The machine running PF is acting as a firewall between a small, internal network and the Internet. Only the filter rules are shown; queueing, nat, rdr, etc., have been left out of this example. 

ext_if  = "fxp0"
int_if  = "dc0"
lan_net = "192.168.0.0/24"

# table containing all IP addresses assigned to the firewall
table <firewall> const { self }

# don't filter on the loopback interface
set skip on lo0

# scrub incoming packets
match in all scrub (no-df)

# setup a default deny policy
block all

# activate spoofing protection for all interfaces
block in quick from urpf-failed

# only allow ssh connections from the local network if it's from the
# trusted computer, 192.168.0.15. use "block return" so that a TCP RST is
# sent to close blocked connections right away. use "quick" so that this
# rule is not overridden by the "pass" rules below.
block return in quick on $int_if proto tcp from ! 192.168.0.15 \
   to $int_if port ssh

# pass all traffic to and from the local network.
# these rules will create state entries due to the default
# "keep state" option which will automatically be applied.
pass in  on $int_if from $lan_net
pass out on $int_if to $lan_net

# pass tcp, udp, and icmp out on the external (Internet) interface. 
# tcp connections will be modulated, udp/icmp will be tracked
# statefully.
pass out on $ext_if proto { tcp udp icmp } all modulate state

# allow ssh connections in on the external interface as long as they're
# NOT destined for the firewall (i.e., they're destined for a machine on
# the local network). log the initial packet so that we can later tell
# who is trying to connect. use the tcp syn proxy to proxy the connection.
# the default flags "S/SA" will automatically be applied to the rule by
# PF.
pass in log on $ext_if proto tcp to ! <firewall> \
   port ssh synproxy state

[Previous: Tables] [Contents] [Next: Network Address Translation]

[back] www@openbsd.org
$OpenBSD: filter.html,v 1.60 2011/11/02 01:01:47 nick Exp $