2.3. Concepts in the nftables framework

Base chains in nftables are highly configurable entry-points in the packet processing pipeline that enable users to specify the following:

This flexibility enables precise control over when and how the rules are applied to packets as they pass through the network stack. A special case of a chain is the route chain, which is used to influence the routing decisions made by the kernel, based on packet headers.

The nftables framework uses an internal virtual machine to process rules. This virtual machine executes instructions that are similar to assembly language operations (loading data into registers, performing comparisons, and so on). Such a mechanism allows for highly flexible and efficient rule processing.

Enhancements in nftables can be introduced as new instructions for that virtual machine. This typically requires a new kernel module and updates to the libnftnl library and the nft command-line utility.

Alternatively, you can introduce new features by combining existing instructions in innovative ways without a need for kernel modifications. The syntax of nftables rules reflects the flexibility of the underlying virtual machine. For example, the rule meta mark set tcp dport map { 22: 1, 80: 2 } sets a packet’s firewall mark to 1 if the TCP destination port is 22, and to 2 if the port is 80. This demonstrates how complex logic can be expressed concisely.

The nftables framework integrates and extends the functionality of the ipset utility, which is used in iptables for bulk matching on IP addresses, ports, other data types and, most importantly, combinations thereof. This integration makes it easier to manage large and dynamic sets of data directly within nftables. Next, nftables natively supports matching packets based on multiple values or ranges for any data type, which enhances its capability to handle complex filtering requirements. With nftables you can manipulate any field within a packet.

In nftables, sets can be either named or anonymous. The named sets can be referenced by multiple rules and modified dynamically. The anonymous sets are defined inline within a rule and are immutable. Sets can contain elements that are combinations of different types, for example IP address and port number pairs. This feature provides greater flexibility in matching complex criteria. To manage sets, the kernel can select the most appropriate backend based on the specific requirements (performance, memory efficiency, and others). Sets can also function as maps with key-value pairs. The value part can be used as data points (values to write into packet headers), or as verdicts or chains to jump to. This enables complex and dynamic rule behaviors, known as "verdict maps".

The structure of nftables rules is straightforward. The conditions and actions are applied sequentially from left to right. This intuitive format simplifies rule creating and troubleshooting.

Conditions in a rule are logically connected (with the AND operator) together, which means that all conditions must be evaluated as "true" for the rule to match. If any condition fails, the evaluation moves to the next rule.

Actions in nftables can be final, such as drop or accept, which stop further rule processing for the packet. Non-terminal actions, such as counter log meta mark set 0x3, perform specific tasks (counting packets, logging, setting a mark, and others), but allow subsequent rules to be evaluated.