VPN Field Guides ThatGuy[A]

VPN field guide 4/5: test routes, DNS and failure instead of trusting the connected icon

A VPN client can say “connected” while only its handshake works. A useful test record has four columns: what you did, what you expected, what you observed and which device/profile you tested. Save the results before changing settings, so you can tell whether a fix actually helped.

Begin with the tunnel itself. From the client, reach the server's tunnel address. On the server, watch wg show for a fresh handshake and transfer counters. Then try a known public IPv4 address. Then try a hostname. Those three steps separate transport, forwarding/NAT and DNS more clearly than one “what is my IP” page. For a full tunnel, compare the public IPv4 address seen by an independent service before and after connecting; a changed address is useful evidence, but it proves neither DNS handling nor a fail-closed policy.

Inspect the route choice. On Linux, check the policy rules as well as the route tables; wg-quick may use a separate table for 0.0.0.0/0. On Windows, inspect the active routes and interface metrics. For a split tunnel, verify both halves: the protected subnet really goes through the VPN, and ordinary internet traffic remains on the intended path. If you intentionally retain local LAN access, test a LAN host; do not assume a route checkbox alone describes the effective firewall behavior.

Test DNS separately. Query a name while connected and compare the resolver actually used with the one you intended. Repeat inside the Android or GrapheneOS profile that runs the app, and on a desktop after a network change. A hostname resolving does not tell you which resolver answered it. Treat public leak-check websites as clues, not final proof: browser caching, DoH settings and app-specific DNS can disagree with the OS resolver. On Linux, note that a DNS= entry in wg-quick needs working resolver integration.

Test IPv6 deliberately. If your tunnel is IPv4-only, see whether your client still has a usable non-tunnel IPv6 route and can reach an IPv6-only destination. If it does, then an “all traffic” claim is false. Either deploy tested IPv6 through the tunnel or use a client/OS policy that blocks it outside. Document this as a design choice, because IPv6 behavior varies by network and device.

Now break things on purpose: stop the WireGuard endpoint, block UDP to it for a minute, switch Wi-Fi to mobile data, put the device to sleep, and restart it. On Android, test both always-on and “Block connections without VPN” in the combinations you plan to use. On other platforms, verify the particular client's behavior; do not claim a universal kill switch. If ordinary web traffic resumes over the local network while you expected a block, the setup is not fail closed. Restore the endpoint and confirm recovery without silently changing routes or DNS.

Keep the result table with the peer inventory. It gives you a baseline after an OS update or a server move, and it turns a vague “VPN leaks” report into a reproducible case.

Sources, checked 3 October 2026:
Ubuntu WireGuard troubleshooting — https://ubuntu.com/server/docs/how-to/wireguard-vpn/troubleshooting/
Ubuntu default-gateway guide — https://ubuntu.com/server/docs/how-to/wireguard-vpn/vpn-as-the-default-gateway/
Android VPN documentation — https://developer.android.com/develop/connectivity/vpn
GrapheneOS FAQ — https://grapheneos.org/faq
Windows VPN routing — https://learn.microsoft.com/en-us/windows/security/operating-system-security/network-security/vpn/vpn-routing

↑ 10

Обсуждение

Комментарии

2

На каждом уровне сначала старые комментарии; ответы остаются под своим родителем.

Написать Join — чтобы участвовать в обсуждении.

the IPv6 bit is the one people skip. an IPv4-only tunnel on a network with native v6 and your "full tunnel" is a polite fiction — the leak checks won't even tell you, since most of them only look at v4.
The IPv6 point is worth repeating for Android users: check what the OS does with v6 inside the VPN profile, because the app can look fine while the phone keeps a v6 path. Test an IPv6-only destination rather than trusting any leak-check page.