Cell-Level Firewalls: Transparent Mode or Routed Mode?

Transparent and routed mode for a cell firewall: the effect on PROFINET DCP, LLDP and multicast, the gateway mistake, the fail-open/bypass decision and the commissioning order on site.
Technical summary
Technology
Industrial firewall, transparent (bridge) mode, routed mode, PROFINET, EtherNet/IP, VRRP, nftables
Topic
Operating mode and placement of the cell/machine firewall
Level
Intermediate–advanced
Prerequisites
  • VLAN and subnet concepts
  • Basic knowledge of PROFINET or EtherNet/IP
  • Experience with firewall rule management
In this article8 sections

The moment a cell firewall is decided on, the first question is: will the device go in as a bridge between two switches without touching the existing IP plan (transparent mode), or will the cell be given its own subnet with the firewall as its gateway (routed mode)? The wrong choice either stops the line or leaves the firewall on paper. Below, from an OT network engineer’s perspective, we describe how the two modes work, their interaction with PROFINET and EtherNet/IP, redundancy and fail-open decisions, and the commissioning order.

What risk does a cell firewall reduce?

A firewall at the zone boundary (for example the DMZ firewall between IT and OT) stops threats from outside; it does not stop a problem spreading inside the OT network. A cell firewall closes this gap: a compromised HMI, a misconfigured engineering laptop or a device generating a broadcast storm in one cell cannot reach other cells and line-level servers. In the field, it does three jobs:

  • Narrowing access: Only SCADA, the historian collector and the authorised engineering workstation access the cell, on specific ports.
  • Restricting functions: On products that support it, reading and writing are separated in protocols such as Modbus and S7; the rule logic is covered in the article Modbus and S7 DPI rules on industrial firewalls.
  • Fault isolation: It limits Layer 2 problems inside the cell (loops, broadcast storms, faulty multicast) from spilling out. How much it limits them depends entirely on the mode.

The cell’s boundaries should be defined in advance with the IEC 62443 zone and conduit approach; the firewall is the enforcement point on that zone’s conduit. The fixed rule of the design: real-time control traffic does not pass through the cell firewall; the controller and I/O devices stay in the same cell. Only supervisory (SCADA/HMI), data collection and engineering traffic passes through the firewall. A design that violates this produces I/O faults every time the firewall restarts.

How transparent mode works and its limits

In transparent mode, the firewall works like a bridge between its two interfaces. The interfaces on the production side have no IP address; the cell and the upstream network stay on the same subnet. While forwarding frames, the firewall performs stateful inspection at the IP layer and, if the product supports it, industrial protocol inspection. Management is done through a separate management interface or management VLAN.

Its biggest advantage is easy commissioning: the IP address, subnet mask or gateway does not change on PLCs, HMIs and drives, and the PLC project is not touched. That is why, in projects where security is added later to a flat network, the decision often goes in favour of transparent mode. The job is done without calling the OEM to site.

The cost, however, is at the architectural level:

  • The broadcast domain is not split. The cell and the upstream network remain in the same broadcast domain. The firewall has to pass ARP; how much it filters non-IP EtherTypes and broadcast/multicast traffic depends on the product. Any Layer 2 traffic that is not filtered can spill out of the cell.
  • The non-IP protocol decision is made explicitly. Frames such as PROFINET RT and DCP (EtherType 0x8892), MRP (0x88E3) and LLDP (0x88CC) do not go through IP inspection; verify whether the firewall passes them by default; do not assume.
  • STP interaction. As an element that forwards or drops BPDUs, the bridge is part of the topology. If there are redundant paths, the firewall’s BPDU behaviour determines the loop risk.
  • The address plan does not improve. The flat network’s IP chaos remains as it is. Transparent mode is a transition tool, not a target architecture.

Advantages and requirements of routed mode

In routed mode, the cell has its own subnet and the firewall is the cell’s default gateway. Layer 2 is physically cut at the cell boundary: broadcast storms, DCP discovery frames or faulty multicast cannot get out. Rules are written legibly per subnet, logs carry meaning and the structure scales as the number of cells grows. In the long term, this is the right approach.

Its requirements, however, are the part most often overlooked in the field:

  • Readdressing: All devices in the cell are moved to the new subnet; IP definitions in the PLC project, HMI connections and SCADA driver configurations are updated. This means a downtime window and, often, the machine builder’s involvement.
  • Gateway definition: On field devices, the gateway field has either been left empty or entered incorrectly; this is the first fault commonly seen when moving to routed mode. A device without a gateway cannot respond to a SCADA outside its own subnet. Either the gateway is fixed on all devices, or source NAT is applied on the firewall for traffic going to the cell; the second way works, but the real source address is lost in the logs.
  • Discovery by engineering tools: DCP-based discovery, such as TIA Portal’s accessible devices scan, runs at Layer 2 and does not cross a routed boundary. The connection is established by entering the IP address; tell the engineering team this in advance, or on day one you will get the complaint “the firewall cut off the PLC”.

Compatibility with industrial protocols: DCP, LLDP and multicast

What most affects the choice of mode is which protocols need to cross the cell boundary. Typical behaviour:

Traffic Transparent mode Routed mode
PROFINET RT (0x8892) Can technically pass; should not Does not pass (Layer 2)
PROFINET DCP discovery Depends on product; should be filtered explicitly Does not pass
LLDP (01:80:C2:00:00:0E) 802.1D bridges usually do not forward it Does not pass
EtherNet/IP implicit I/O (UDP 2222, multicast) Can pass; risk of spreading Requires multicast routing; unicast preferred
Modbus TCP, S7comm, OPC UA (TCP) Passes with stateful inspection Passes with stateful inspection

Because bridges do not forward LLDP, the firewall can appear as a “missing neighbour” in PROFINET topology diagnostics; update the topology definition inside the cell so that it ends at the firewall port. On the EtherNet/IP side, if there is a producer–consumer connection between cells, first question whether it is really needed; if it is and the devices support it, switch to a unicast implicit connection. How multicast is contained within the cell is covered in the article EtherNet/IP multicast and IGMP snooping.

In transparent mode, non-IP traffic must be handled explicitly. The example below was written with the Linux nftables bridge family to show the vendor-independent logic; on a flat 192.168.10.0/24 network, the cell PLCs are in the .50–.59 range and SCADA is at .5:

table bridge cell_fw {
  chain forward {
    type filter hook forward priority 0; policy drop;
    ct state established,related accept
    ether type arp accept
    ether type 0x8892 drop    # PROFINET RT/DCP does not leave the cell
    ether type 0x88e3 drop    # MRP frames stay in the cell
    ip saddr 192.168.10.5 ip daddr 192.168.10.50-192.168.10.59 tcp dport 102 ct state new accept
    ip saddr 192.168.10.5 ip daddr 192.168.10.50-192.168.10.59 tcp dport 502 ct state new accept
  }
}

On commercial industrial firewalls, the same decision is made with “non-IP protocol” or “Layer 2 filter” settings. Verify the default behaviour with a packet capture on site, not from the documentation. For syntax details, the nftables wiki is the primary source.

Redundancy, fail-open and bypass decisions

A cell firewall enters the design as a single point of failure. First answer this question: if the firewall stops, what is lost in the cell? If control traffic stays inside the cell, the machine keeps running; what is lost is SCADA visibility, data collection and remote intervention. If the HMI is outside the cell, the operator cannot run the machine. In that case, bringing the HMI inside the cell is a simpler and more robust solution than firewall redundancy.

  • Fail-closed: Traffic is cut if the firewall fails. This is the preferred default from a security standpoint; if in-cell control communication does not pass through the firewall, it is unaffected by this failure, while data flows leaving the cell are cut.
  • Fail-open / hardware bypass: Some industrial firewalls have bypass relays that physically connect the ports together when power is lost. This only makes sense in transparent mode (in routed mode, bypass joins two different subnets but provides no routing). The moment bypass engages, security control disappears completely; this state must be monitored as an alarm.
  • Active/passive pair: In routed mode, a virtual IP is used as the gateway (VRRP or the manufacturer’s HA mechanism). In transparent pairs, both devices forwarding at the same time creates a Layer 2 loop; do not deviate from the manufacturer’s HA design guide.

If the cell is an MRP ring, the firewall is placed not inside the ring but at the uplink point where the ring connects to the upstream network. For the ring structure, see the article ring topology with PROFINET MRP.

Comparison table and decision criteria

Criterion Transparent Routed
IP change None Required
Layer 2 isolation Partial, depends on product Full
Commissioning downtime Short (inserting into the cable) As long as readdressing takes
Rule readability Host-based lists Subnet-based, simple
Bypass option Meaningful Not meaningful
Long-term target architecture Transition solution Suitable

A common approach: in an existing cell that is in production and where IP changes require OEM approval, start with transparent mode; move the cell to routed mode at the next major overhaul or during a downtime suitable for readdressing. In new line and new machine projects, consider a routed design from the start and write the decision into the purchase specification. The order in which this decision is applied when splitting a flat network cell by cell is part of the same migration plan.

Commissioning steps

Building the traffic baseline and communication matrix is described in detail in the article migrating a flat OT network to segmentation; here we cover only the steps specific to the cell firewall:

  1. Baseline and matrix: From a SPAN/TAP capture on the cell uplink, in addition to IP flows, list separately the non-IP EtherTypes (0x8892, 0x88E3, 0x88CC) and BPDUs crossing the cell boundary. This list decides the mode.
  2. Placement and cabling: Mount the firewall between the cell switch and the distribution switch, in the same cabinet if possible. Connect the management interface to the management network.
  3. Monitoring mode: Run it with a policy that allows all traffic but logs it. Catch unexpected non-IP traffic and STP behaviour at this stage.
  4. Rule base: Write the rules from the communication matrix on the principle of least privilege; the last rule is “deny + log”.
  5. Enforcement and testing: Switch to enforcement mode during planned downtime; test in turn the SCADA screens, alarm flow, historian collection, online access from the engineering workstation and the remote support scenario. Write down the step for returning to monitoring mode in advance.
  6. Documentation: Keep the rule justifications, communication matrix and firewall configuration backup under version control together with the PLC project.

For defining cell boundaries and the communication matrix, see our OT cybersecurity and segmentation page, and for the physical design of the cell network, our industrial Ethernet and factory networks page. Once the flows between cells are defined, verify the access rules and their limits in an IT/OT security segmentation and access control engagement.

Frequently asked questions

Is a firewall in transparent mode never given an IP address?

The bridge interfaces on the production side are not given an IP. An IP address on a separate management interface or management VLAN is used for management, log forwarding and time synchronisation; this address must not be on the production subnet.

What happens if the controller is outside the firewall in a PROFINET cell?

PROFINET RT is Layer 2 traffic and cannot pass through a firewall in routed mode; even if it passes in transparent mode, cycle time and stability become dependent on the firewall. The controller and I/O devices stay in the same cell, and only supervisory and data collection traffic passes through the firewall.

Should there be a separate firewall in front of each machine, or one firewall per cell?

The decision depends on the zone definition. Machines with the same risk profile that communicate heavily with each other can be grouped into a single cell and protected with one firewall. For machines from different manufacturers, with remote access, or with a different target security level, a separate firewall is more appropriate.

What can I do if I cannot change the PLCs’ gateway when moving to routed mode?

Source NAT can be applied on the firewall for traffic going to the cell; the PLC sees the SCADA request as if it came from the firewall’s address on its own subnet and does not need a gateway. It works, but the real source does not show up in logs on the PLC side, and rule management becomes more complex. The lasting solution is to fix the gateway.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Industrial

PLC, PROFINET/PROFIBUS and SCADA communication, IT/OT network segmentation and industrial field connectivity. We install and monitor without stopping the production line.

Knowledge Center

Other articles