Skip to content

OT Network Segmentation and the Industrial DMZ

Design OT network segmentation that holds up: zones, an industrial DMZ, adjacent-level firewall rules, ICS protocol allow lists, data diodes and safe rollout.

Published on 8 min read

Segmentation is the single control that shows up in every OT incident report's recommendations. The 2015 Ukraine grid attack, the 2016 Industroyer event and many ransomware outages since all involved adversaries or malware moving from business networks into control networks, as described in our ICS attack case studies. Good segmentation does not stop every attack, but it turns a flat network where one compromised laptop reaches every PLC into a set of small, monitored zones with narrow, known paths between them.

This guide covers how to design zones, build an industrial DMZ (IDMZ), write firewall rules for ICS protocols, and roll changes out without stopping production.

Principles first

Three principles from the standards drive the design:

  1. Group assets into zones with the same security requirements, and control every communication path between them through conduits. This is the core of IEC 62443 foundational requirement 5, Restricted Data Flow.
  2. Only allow connections between adjacent levels. NIST SP 800-82 Rev. 3 states that, in a Purdue architecture, Level 4 devices should not communicate directly with Level 2, 1 or 0 devices.
  3. Treat outbound traffic from the control network as strictly as inbound. NIST calls out unmanaged outbound connections from lower levels as a significant risk. Malware command and control and data exfiltration both travel outbound.

Add a practical fourth: no traffic crosses between the enterprise and the control network directly. Everything terminates in the IDMZ.

A reference zone layout

Most plants map zones onto the Purdue model because the vocabulary is shared by engineers and IT:

ZoneTypical contentsTalks to
Enterprise (L4 and L5)ERP, email, corporate AD, internetIDMZ only
IDMZ (L3.5)Historian replica, patch and AV staging, remote access gateway, file transferEnterprise and site operations, never through
Site operations (L3)Site historian, engineering workstations, OT domain controller, backup serverIDMZ, supervisory zones
Supervisory (L2), one per areaHMI, SCADA servers, alarm serversSite operations, its own control zone
Control (L1), one per areaPLCs, DCS controllers, RTUsIts own supervisory zone, its I/O
SafetySIS controllers and SIS engineering stationNarrow, read-only conduit to BPCS

Split Level 2 and Level 1 by process area or production line, so that a problem on line 1 cannot reach line 2. Keep the safety system in its own zone with the narrowest conduit in the plant.

Designing the industrial DMZ

The IDMZ is a buffer network between the enterprise and site operations. Its rule is simple: a connection from either side ends at a server in the IDMZ. Data passes through by being re-published, not routed.

Common IDMZ services:

  • Historian replication: the site historian (L3) pushes data to a replica in the IDMZ; enterprise users query the replica. Nobody in the enterprise queries the plant historian directly.
  • Patch and antivirus staging: updates are downloaded on the enterprise side, validated, staged in the IDMZ, then pulled by an OT update server at Level 3. See OT vulnerability management.
  • Remote access broker: users and vendors authenticate to a gateway in the IDMZ, then reach a jump host in Level 3 through a separate, monitored session. Details in secure remote access for OT vendors.
  • File transfer: a controlled file exchange point with malware scanning, instead of shared drives that span both networks.

Design rules for the IDMZ:

  • Use two firewall layers (or a single firewall with separate, strictly filtered interfaces) so that compromise of one rule set does not expose Level 3.
  • Do not extend the enterprise Active Directory into OT. Use a separate OT domain (or local accounts) and avoid trusts that allow enterprise credentials to log on in OT.
  • No dual-homed hosts. A server with one NIC in the enterprise and one in OT bypasses every firewall. NIST SP 800-82 also warns against dual-homed connections.
  • Log every connection crossing the IDMZ, and send those logs to a collector both IT and OT security can read.

Firewall rules for ICS protocols

Start from a default deny in both directions and add explicit, documented allows built from observed and approved flows. Your passive asset inventory gives you the communicating pairs; turn each approved pair into a rule.

Common ICS ports from the IANA registry:

ProtocolPort
Modbus/TCP502/tcp (Modbus Secure: 802/tcp)
S7comm, IEC 61850 MMS (ISO-TSAP)102/tcp
EtherNet/IP explicit messaging44818/tcp and udp
EtherNet/IP implicit I/O2222/udp (EtherNet/IP over TLS/DTLS: 2221)
DNP320000/tcp and udp
IEC 60870-5-1042404/tcp
OPC UA4840/tcp (OPC UA over TLS: 4843)
BACnet/IP47808/udp

An example nftables policy for a Linux-based boundary between site operations (L3) and a supervisory zone (L2):

table inet ot_boundary {
  set hist_collectors { type ipv4_addr; elements = { 10.30.1.10 } }
  set hmis            { type ipv4_addr; elements = { 10.20.1.21, 10.20.1.22 } }
  set eng_ws          { type ipv4_addr; elements = { 10.30.1.50 } }
  set l1_plcs         { type ipv4_addr; elements = { 10.10.1.11, 10.10.1.12 } }

  chain forward {
    type filter hook forward priority 0; policy drop;
    ct state established,related accept
    ct state invalid drop

    # L3 historian collector reads from L2 OPC UA server
    ip saddr @hist_collectors ip daddr 10.20.1.30 tcp dport 4840 accept

    # L2 HMIs poll L1 PLCs over Modbus/TCP
    ip saddr @hmis ip daddr @l1_plcs tcp dport 502 accept

    # Engineering workstation programs PLCs: S7 (102) and EtherNet/IP (44818)
    ip saddr @eng_ws ip daddr @l1_plcs tcp dport { 102, 44818 } accept

    # Log everything else before the policy drops it
    log prefix "ot-boundary-drop " counter
  }
}

The same logic applies to any commercial firewall:

  • Name objects by function (HMI_LINE1, PLC_LINE1) so rules are reviewable by engineers.
  • Use explicit source and destination hosts, never whole subnets to whole subnets for write-capable protocols.
  • Separate read paths (historian, HMI polling) from write and program paths (engineering workstations). Only the engineering workstation should reach controller programming services.
  • Log denies, and review them during commissioning: they reveal flows nobody documented.

Going deeper than ports

Port-based rules cannot tell a Modbus read (function code 3) from a write (function code 16) or a firmware update. Industrial firewalls with deep packet inspection can enforce at the function code level, for example allowing a historian to use only Modbus read functions 1 to 4 against a PLC. Use this on conduits where a read-only flow is expected, especially toward safety systems. Protocol details are in ICS protocol security.

Unidirectional gateways

For flows that only need to leave the control network (historian data, alarm feeds, log forwarding), a unidirectional gateway (data diode) enforces one-way transfer in hardware. NIST SP 800-82 notes that a gateway placed between Levels 2 and 3 can protect Levels 0 to 2 from events at Levels 3 and above. Trade-offs: replication software is needed on both sides, and anything bidirectional (remote access, patch delivery, acknowledgements) needs a separate, controlled path or a different design.

VLANs are not a security boundary on their own

VLANs are a good way to organize traffic inside a zone, but VLAN separation alone does not filter traffic between zones. If inter-VLAN routing is enabled on a core switch without ACLs, every VLAN can reach every other. Enforce zone boundaries with firewalls or ACLs at the routing point, disable unused switch ports, and turn off trunk negotiation on access ports.

Rolling out segmentation without breaking production

Most outages caused by segmentation projects come from an undocumented flow that someone needed. A safe rollout:

  1. Monitor first. Capture traffic at the planned boundary for at least a full production cycle (shifts, batches, weekly reports, backups).
  2. Draft rules from observed flows, then review each one with operations and the control vendor. Remove flows nobody can justify.
  3. Deploy in monitor mode where the firewall logs but does not block (or with a final allow-and-log rule) and compare logs with the draft rule set.
  4. Switch to enforcement during a maintenance window, with engineers on site and a tested rollback.
  5. Keep the deny log under review for the next few weeks, and add any legitimate flows through change management, not ad hoc exceptions.

Validating the result

Segmentation should be tested, not assumed:

  • From an enterprise host, confirm you cannot reach Level 3 or below: only IDMZ services should answer.
  • From a Level 3 host, confirm only approved services on approved controllers are reachable.
  • Check for bypasses: dual-homed hosts, cellular modems in cabinets, vendor remote access boxes, wireless bridges and engineering laptops that connect to both networks.
  • Map the result against MITRE ATT&CK for ICS mitigations Network Segmentation (M0930) and Filter Network Traffic (M0937), and review it whenever the plant changes.

Segmentation also shapes OT incident response: well-defined conduits are the places where you can isolate a zone during an incident without stopping the whole site. Plan which conduits can be cut, by whom, and what the process does when they are.

References

Related articles

Design vendor remote access to OT you can approve, watch and revoke: brokered sessions, DMZ jump hosts, MFA, time limits and firewall rules to enforce it.
Build an OT incident response plan around safety and process continuity: roles, decision rights, playbooks, evidence from PLCs and HMIs, and verified recovery.
Run OT vulnerability management that works: read CISA ICS advisories, pull CSAF data, check KEV, rank by exposure and consequence, then patch or compensate.