On many internal networks, a live jack left in a meeting room wall or a free port in a switch cabinet on the shop floor is enough: whoever plugs something in gets an IP address. By default the switch does not ask who has connected. It simply forwards the frames.
802.1X asks that question at the port. Until the device has proven its identity, the port passes only the traffic needed for authentication, and after that a central server decides which network the device lands in and what it can reach. Below we go through the roles and the authentication methods, look at how devices that do not support the standard get onto the network, and explain how to turn on enforcement without leaving anyone offline along the way.
Supplicant, authenticator, RADIUS server
The IEEE 802.1X standard defines three roles. The supplicant is the software on the connecting device that presents the credentials; it is built into today's desktop and mobile operating systems. The authenticator is the switch or wireless access point whose port the device connects to. The authentication server makes the decision, and in practice it is almost always a RADIUS server backed by a directory, a certificate authority, or a device database.
The steps of one authentication
- The port comes up. At this point the switch passes only EAPOL frames (EAP over LAN) and drops all other traffic.
- The switch asks the device for an identity, and the supplicant answers, for example with a username or the name in its certificate.
- The switch wraps the EAP message in a RADIUS request and sends it to the server. From here on it acts as a relay: the supplicant and the server carry out the authentication between themselves, and the switch never sees the password or the certificate's private key.
- The server sends back an accept or a reject. On accept, the switch opens the port to the device's traffic.
What the server can send back
An accept message can also carry attributes. The best known is dynamic VLAN assignment: the server tells the switch which VLAN to put the port in, so the same wall jack places a workstation on the office network, an IP phone on the voice network, and a camera on the camera network. Depending on the switch model, the response can also include an access list, a session timeout, or a reauthentication interval.
The decision can also be changed afterward. With a RADIUS Change of Authorization (CoA) message, the server can ask the switch to reauthenticate the device, apply a different rule to it, or drop the session. You need this when something about the device comes to light later, for example its profile changes or a security system raises an alert about it.
How many devices can share a port
The port's host mode determines how the switch handles more than one MAC address showing up on a port. In single-host mode, only one authenticated device can be on the port. In multi-domain mode, the IP phone and the computer connected behind it authenticate separately, with the phone placed in the voice VLAN and the computer in the data VLAN. In a mode that serves multiple devices, each device authenticates on its own. There is also a setting in which the first successful authentication opens the whole port. It is convenient, but anyone can get in through a small switch plugged in behind the first device, so use it only when there is a real reason to.
EAP methods: certificate or password
EAP (Extensible Authentication Protocol) is only a framework. What the device uses to prove its identity is set by the EAP method you choose, and the security of 802.1X depends largely on that choice.
EAP-TLS
With EAP-TLS, both sides prove their identity with a certificate: the server to the client and the client to the server. There is no password, so there is nothing to steal with a phishing email or a fake login page. The price is a working PKI. Every device needs its own certificate, which has to be issued automatically, renewed before it expires, and revoked when the machine is retired. If the RADIUS server cannot reach the revocation list or the OCSP responder, then depending on the configuration (with permissive, so-called soft-fail checking) a revoked certificate can still get through, so this chain needs checking too.
Tunneled methods with a password
PEAP and EAP-TTLS first build a TLS tunnel in which only the server presents a certificate, and the username and password are checked inside that tunnel, with PEAP typically using MSCHAPv2. Their advantage is that they work with existing directory accounts and need no client-side certificate.
The weak point is server certificate validation. If the client accepts any certificate, a rogue access point or RADIUS server can run the login with it, and the password can be recovered from the captured MSCHAPv2 response by offline guessing. So the client has to be pinned to the certificate authority and server name it accepts, and users should not be able to click through the warning.
Machine and user
A domain-joined workstation needs the network before anyone logs on, so it can reach the directory and download policies. That is what machine authentication is for. It runs with the computer's certificate or account, and once the user signs in, user authentication can follow. TEAP can do both inside a single tunnel, so the server can also see whether that user is working on a company machine.
| Method | How the client proves its identity | Prerequisite | Typical weak point |
|---|---|---|---|
| EAP-TLS | Its own certificate and private key | PKI, automatic certificate enrollment | Expired certificates, revocation not checked |
| PEAP (MSCHAPv2) | Username and password inside the TLS tunnel | Server certificate, directory | Server certificate validation turned off on the client |
| EAP-TTLS | Inner authentication inside the TLS tunnel | Server certificate, a client that supports it | Same as PEAP |
| TEAP | Certificate, password, or both chained | Server certificate, newer client-side support | No support on older systems |
EAP-MD5 is still in the base EAP specification (RFC 3748), but it provides neither mutual authentication nor an encryption key, so it has no place on an enterprise network. Newer editions of 802.1X can also use the key derived during authentication to encrypt the wired link (MACsec), which requires support on both the switch and the client.
MAB for devices without 802.1X support
Every network has devices with no supplicant: network printers, IP cameras, access control panels, building management modules, older phones, and in OT environments most of the controllers and field devices. For these, what is left is MAC address-based authentication (MAC Authentication Bypass, MAB).
It works simply. The switch starts authentication on the port, and if no EAPOL response arrives after a few attempts, it reads the source MAC address from the device's first frame. It sends this address to the RADIUS server as the username, and the server checks whether it is on the list of allowed devices and, if so, what access it gets. There is no cryptographic proof here. The server sees only an address.
A MAC address is not a secret
The MAC address is printed on the device label and sent in every frame, and on most operating systems a single setting changes it. Anyone who unplugs a hallway printer and plugs in a laptop set to the printer's address gets the printer's access. MAB is therefore only ever partial identification, and it has to be offset with narrow access rights:
- Each device type should reach only what it needs: the printer the print server, the camera the recorder. An access list or firewall rule should block every other destination.
- Profiling should keep checking that the same kind of device is still behind the address. If a “printer” suddenly shows up with a workstation's DHCP fingerprint, that warrants an alert and revoked access.
- For known devices in fixed locations, the address can be bound to a switch or a port. A link going down on such a port is itself an event worth watching.
- Review the list of allowed addresses regularly, because entries for retired devices do not go away on their own.
Order and timing
By default the switch tries 802.1X first and falls back to MAB only after that fails. For MAB devices this means a delay, and during it some devices' DHCP requests or network boot time out. The reverse order is faster, but then you have to configure the switch so that an 802.1X response arriving later overrides the MAB result. On OT devices, reauthentication is a risk as well: a brief interruption or an unreachable RADIUS server can stop a production process, so the reauthentication interval and the behavior during a server outage need separate planning on these ports.
Profiling: how to recognize a printer or a camera
Profiling tries to work out the device type from what is visible about it on the network, so that the MAB decision does not rest on a hand-maintained address list. A single source is rarely enough, so the profiler cross-checks several signals.
- OUI. The IEEE assigns the first part of the MAC address (typically the first three bytes) to manufacturers, so the address tells you who made the network adapter. That alone does not tell you the device type, and phones and laptops on Wi-Fi often use a random, locally administered address.
- DHCP. In the device's DHCP request, the list and order of requested options (option 55) are characteristic of the operating system and even the firmware. On top of that come the vendor class identifier (option 60) and the hostname (option 12). The profiler receives these from a copy sent by the DHCP relay or from device data sent by the switch.
- LLDP. Devices that speak LLDP, or the LLDP-MED variant common on phones, announce their name and capabilities (for example phone or bridge), and often their model and software version too.
- Web traffic. The User-Agent header in HTTP requests, or a visit to a registration portal, shows the browser and the operating system.
- Active queries. SNMP queries, open port checks (a printer typically accepts print jobs on port 9100, a camera streams video over RTSP), and mDNS or NetBIOS responses.
Profile rules compute a score or certainty level from these signals and grant access only above a threshold. If the device's profile changes later, the server forces a new decision with a Change of Authorization message.
An isolated zone for unknown devices
Authentication rarely has only two outcomes. In practice several situations come up, and each one needs its own decision:
- Unknown device. It has no supplicant, and its address is not on the list. It typically goes into a guest or registration zone, from which only the internet or a registration portal is reachable.
- Failed 802.1X. A wrong password, or an expired or revoked certificate. These devices are best placed in a separate zone where the services needed to renew the certificate or reach the helpdesk are available, but the internal systems are not.
- Noncompliant device. If network access control also checks device posture (a missing update, endpoint protection turned off), a failing machine can be moved to a remediation zone from which only the update servers are reachable. This usually needs agent software on the device.
- Unreachable RADIUS server. This takes an operational decision: should the ports stay closed, or should they let devices through in a predefined VLAN? In an office the first option may be acceptable. On a production OT network the second is almost always the right one.
A separate VLAN number is not enough for an isolated zone. It also needs a firewall or access rule that blocks all traffic toward the internal networks and allows only the required DHCP, DNS, and portal traffic. How the rules between segments are built is covered in our article on segmenting the internal network.
VLAN changes often catch people out. If a device has already received an IP address and the server then moves it to another VLAN, the device keeps trying with its old address until the lease expires. So a Change of Authorization should also request a brief port bounce, which makes the device ask for a new address.
Monitor mode and phased enforcement
If 802.1X goes into enforcement on every port on the same day, every forgotten printer, misconfigured client, and expired certificate turns into a ticket at once. That is why a rollout usually starts in monitor mode: the port runs authentication and the server logs the result, but devices that fail still get access. The log shows who would fail under enforcement, and the first version of the MAB list and the profile rules is built from it.
- Monitoring. Authentication on every access port with nothing blocked, for at least one full business cycle, so that rarely used devices show up as well.
- Fixing. Configuring supplicants through central policy, issuing missing certificates, and adding and profiling MAB devices.
- Restricted mode. Before authentication, the port allows only the traffic permitted by a narrow access list (DHCP, DNS, access to the directory), and grants full access after successful authentication.
- Enforcement. Blocking is switched on site by site or switch group by switch group, starting with the office network. The OT network comes last, in an agreed maintenance window.
The operations work that remains after enforcement
- At least two RADIUS servers with failover configured on the switches, and regularly tested behavior for when neither of them responds.
- A clear process for new devices: who can add them to the MAB list, with what data, and who approves it.
- Central collection of RADIUS logs and accounting records, for example in a SIEM. They show which device appeared on which port and when, and that counts for a lot when investigating an incident.
- Monitoring certificate expiry. When a forgotten CA or server certificate expires, it can stop every certificate-based login (EAP-TLS, PEAP, EAP-TTLS, TEAP) at the same time.
Wired 802.1X relies on the same roles and EAP methods as enterprise Wi-Fi, so the two can run on a shared RADIUS server with shared rules. The specifics of the wireless side are covered in our article on enterprise Wi-Fi security. Device-level identification is also one of the foundations of the Zero Trust approach, where the network likewise grants no trust just because someone plugged into an internal jack.
Related service: Network access control (NAC) →