Troubleshooting
Run a remote access VPN from your VM with NAT mode
Short answer
Set the VM's network adapter to NAT and the guest adapter to DHCP, because NAT translates guest traffic through the host's routing table, including the routes a host-side VPN client adds. Bridged mode fails because the guest appears on the plant LAN as its own device and never sees those routes. Verify by pinging a remote site device from the guest.
On this page
- Why can't my VM reach the plant through the host's remote access VPN?
- Which network adapter mode carries guest traffic through a host VPN?
- How do I switch the VM to NAT and set the guest to DHCP?
- Why does bridged mode fail when the host VPN tunnel is up?
- How do I verify the traffic actually rides the host tunnel?
- What changes when the VPN client uses split tunneling?
- What else breaks a VPN connection from inside a VM?
- Questions asked
Set the virtual machine's network adapter to NAT and the guest operating system's adapter to DHCP — that pair of settings carries guest traffic through a remote access VPN client running on the host. In bridged mode the guest shows up on the plant LAN as its own device, and the tunnel routes the VPN client writes into the host's routing table never reach it. The combination was reported working on VMware Workstation Pro 16, and the routing logic behind it holds on any hypervisor: NAT makes the guest a tenant of the host's network stack, so everything the host can reach through the tunnel, the guest can reach too. The sections below cover the adapter switch, the DHCP detail that silently breaks it, a route check with real numbers, and the split-tunnel limit no adapter setting can bypass.
Why can't my VM reach the plant through the host's remote access VPN? #
The guest is sending traffic on its own path instead of through the host's tunnel: with a bridged adapter the virtual machine appears on the physical LAN as an independent device, and the routes the host VPN client added never apply to its separate stack.
The pattern is distinctive. On the host, the VPN client connects to its remote access service and reports the tunnel up, while the engineering software inside the guest cannot browse to the PLC, download a project, or open the HMI server at the site. Nothing is wrong with the tunnel — the guest is simply not on it. A bridged adapter gives the guest a second, independent network stack with its own routing table, and the VPN client only writes routes where it runs: on the host. The guest keeps handing packets to a plant LAN that has no path to the far end.
Which network adapter mode carries guest traffic through a host VPN? #
Use NAT mode — the adapter setting that shares the host's IP address — because it translates the guest's outbound packets through the host's own address, so every packet follows the host's routing table, including the tunnel routes the VPN client adds, with no second VPN client inside the guest.
NAT makes the guest a tenant of the host's network stack. Outbound packets leave the guest on the NAT subnet, have their source address rewritten to the host's, and are then routed by exactly the table the VPN client edits — tunnel routes included. That is also why no VPN client belongs inside the guest: one tunnel, one route writer, on the host.
The split is not specific to VMware Workstation Pro 16, where the NAT plus DHCP combination was reported working. On macOS virtualization frontends such as UTM, the hardware-backed network modes use the platform's virtualization features and may not respect the host's VPN settings, and the project's own tracker points to its emulated network mode as the workaround. Wherever a mode detaches the guest from the host stack, the tunnel is unreachable from the guest. The same adapter setting is also the fix when a VM cannot open a connection to a controller at all — fixing a VM remote PLC connection with a NAT adapter works that fault from the controller side.
How do I switch the VM to NAT and set the guest to DHCP? #
Shut the guest down, set its virtual network adapter to NAT in the hypervisor's VM settings, boot it, then set the guest adapter to obtain an IP address automatically so the virtual DHCP server puts it on the NAT subnet with the NAT device as its default gateway.
The steps below are written for VMware Workstation Pro 16; every desktop hypervisor exposes the same two choices, adapter mode and guest addressing:
- Shut the guest operating system down completely.
- In the VM settings, open the network adapter and change its connection from Bridged to NAT — the option described as sharing the host's IP address.
- Start the guest, open its own network adapter properties, and set it to obtain its IP address and DNS settings automatically (DHCP).
- Confirm the guest leased an address on the NAT subnet — a different subnet from the plant LAN's — with the NAT device as the lease's gateway.
- Start the remote access VPN client on the host and wait until it reports the tunnel fully up.
- From the guest, ping the address of a device on the remote site network, such as the PLC or the site gateway.
A reply at step 6 proves the chain end to end. No reply means the next two checks — the route behind the tunnel, then the residual causes below — sort out which link failed.
Why does bridged mode fail when the host VPN tunnel is up? #
Bridged mode gives the guest its own MAC address and its own IP address on the physical LAN, so it becomes a separate device that makes its own routing decisions; the tunnel routes exist only in the host's table, and the guest never reads them.
NAT and bridged differ in whose address the network sees. Under NAT, the plant LAN, the VPN service's gateway, and the tunnel all see traffic sourced from the host's address — the same identity that holds the tunnel routes, so the path is consistent end to end. Under bridging, packets leave the guest carrying the guest's own source address and MAC, the host's routing never engages, and the tunnel never sees them. Bridging also puts a second live device on the control network, with its own DHCP lease and its own exposure, which the site's segmentation policy then has to account for.
For the sibling fault where a bridged guest cannot reach the LAN at all, not just the tunnel, this troubleshooting guide for a VMware bridged guest that cannot ping goes deeper on that failure.
How do I verify the traffic actually rides the host tunnel? #
Check the guest's default gateway and then the destination's longest match: the guest's gateway must be the NAT device, and the remote address must fall inside a subnet that the host's tunnel routes cover — both checks take a minute at a command prompt.
Worked example — illustrative values, since each hypervisor assigns its own NAT subnet and each VPN service publishes its own tunnel routes:
- The virtual DHCP server leases the guest 192.168.137.20/24, gateway 192.168.137.2 (the NAT device).
- The remote PLC is at 10.42.3.15.
- On connect, the host VPN client installs the site subnet 10.42.3.0/24 through the tunnel interface.
The guest checks its own table for 10.42.3.15, finds no more specific match, and sends the packet to its default gateway 192.168.137.2 — the NAT device. NAT rewrites the source from 192.168.137.20 to the host's address. On the host, the match test is a bitwise AND: 10.42.3.15 AND 255.255.255.0 = 10.42.3.0 (10 AND 255 = 10, 42 AND 255 = 42, 3 AND 255 = 3, 15 AND 0 = 0), which is exactly the network of the tunnel route 10.42.3.0/24, so the packet leaves through the tunnel.
A reply proves the chain guest to NAT to host routing table to tunnel to site, and back. A trace from the guest should show 192.168.137.2 as the first hop; if the plant router answers first, the guest is not behind NAT. Cloud platforms expose a virtual NIC's effective routes for exactly this diagnosis; a desktop hypervisor gives the same answer from the guest's own table.
What changes when the VPN client uses split tunneling? #
Split tunneling narrows what the guest can reach: only the subnets listed in the tunnel routes pass through the VPN, everything else follows the host's ordinary internet route, so a guest can browse the web but still not see the PLC when its subnet is missing from the tunnel.
NAT still does its part, but it can only forward what the tunnel carries. If the site subnet is missing from the tunnel's route list, the host itself has no path to 10.42.3.0/24, and no guest-side setting can invent one; the fix lives in the tunnel definition on the remote access service side, where the site's subnets are pushed as routes. The two-minute test: from the guest, ping one address inside a listed tunnel subnet and one outside it — the first should answer, the second should not, which proves the adapter is fine and the route list is the constraint. Full-tunnel clients invert this: everything follows the tunnel, so the guest's ordinary internet traffic rides the site path too, with the bandwidth and policy weight that brings.
The tunnel is one leg of the remote path; remote SCADA access through firewalls and proxies covers the controls around the rest of it. One safety note: a remote access tunnel into a site is a door into the OT network — keep the client on a hardened host, keep the pushed route list as narrow as the work needs, and drop the session when the work is done.
What else breaks a VPN connection from inside a VM? #
Three causes cover most of the remaining failures: the host tunnel dropped or never fully connected, the guest holds a static address outside the NAT subnet, or a second VPN client inside the guest is fighting the host's client for the routes.
Work down the short list before touching the adapter again:
- Tunnel state on the host: the client must report the session up before you blame the guest.
- Guest address: confirm the adapter holds a leased NAT-subnet address, not a static leftover from the bridged days pointing at the plant router.
- Second VPN client: uninstall any VPN client running inside the guest.
- Hypervisor mode: on macOS virtualization frontends, use the emulated NAT-style mode, because the hardware-backed modes may ignore the host's VPN settings.
Running the engineering VM on a Linux hypervisor instead of a desktop one changes the mechanics, not the principle: keep the guest behind the host. Proxmox for SCADA and PLC engineering virtual machines covers that host side.
Decision, case by case. VPN client on the host: NAT adapter plus guest DHCP. VPN client inside the guest: bridged is fine, because the routes are written where the guest lives. Both correct and the site is still dark: the tunnel's route list is the fault, not the adapter.
Questions asked
Does the remote access VPN client need to be installed inside the guest too?
No, and adding one is a common way to break a working setup. The guest reaches the site through the single tunnel on the host; a second client inside the guest writes its own competing routes on its own virtual adapter and can blackhole the path entirely. Keep exactly one VPN client, on the host, and let NAT carry the guest's packets through it.
Can the guest keep a static IP address when the VM runs in NAT mode?
Yes, but only with values the virtual DHCP server would have assigned: an address inside the NAT subnet, the NAT device as default gateway, and DNS that resolves through the host. A static address copied from the plant LAN sends guest packets to the plant router, they never reach the NAT device, and the tunnel is bypassed completely.
Does the same NAT pattern hold on other hypervisor platforms?
Yes. Any placement that keeps the guest behind the host inherits the host's routes: Hyper-V's default virtual switch puts the guest behind the host under NAT, and on Linux hosts a NAT (masquerade) rule gives guests the same path. Only modes that attach the guest directly to the physical network take it off the host's routing table and lose the tunnel.
What happens to the guest when the host's VPN session drops?
Its route to the site disappears with the host's. Because NAT makes the guest a tenant of the host's routing table, a dropped tunnel, a host sleep event, or an account disabled by a service plan change cuts the guest off at the same moment. Nothing inside the guest needs reconfiguring; when the tunnel returns, the routes return with it.