Start your 3-day free trial
Sign up to experience all premium features at no cost.
*Available only to new users. Each user is limited to one trial.


When a WireGuard handshake succeeds but no data moves, the peers have authenticated recently, yet that fact does not show what happened to your test packet. Generate one controlled flow and compare fresh transmit and receive deltas on both peers. The resulting pattern tells you whether to inspect local route and peer selection, the outer path, remote forwarding and return routing, or a later layer such as DNS or MTU.
The complete VPN guide covers every connection stage. This article begins only after latest handshake has advanced for the peer you intend to test.
The wg and wg-quick guidance here applies to Linux deployments using those tools. Record your actual OS, kernel/core, and tool versions first; other clients and userspace implementations need their documented equivalents. Gateway changes require administrator permission and a saved recovery path. This guide was checked against the cited documentation; the author has not performed a complete end-to-end test of this procedure.
Key Takeaways
- A current handshake authenticates the peers; it does not prove that a chosen application flow uses the tunnel.[1]
- Compare counter deltas around one test, never lifetime totals from unrelated traffic.
- No local TX points first to local route or peer selection; TX without matching RX leaves remote acceptance unconfirmed.
- Once TX and RX both grow for the probe, move diagnosis to DNS, MTU, transport, or the application.
- Keep private keys, preshared keys, full configurations, and raw diagnostic dumps out of shared evidence.
WireGuard's handshake establishes fresh symmetric session keys between peers that know the expected static public keys. Transport data messages are a separate message type and follow after that authenticated exchange.[1] A current handshake is therefore strong identity and reachability evidence for that moment, but it is not a synthetic transaction through your intended destination.
The wg interface exposes a peer's latest handshake and cumulative transfer bytes.[2] Those counters are useful only when you note them before and after a narrow probe. A background keepalive, another application, or an earlier transfer can make nonzero totals look reassuring even while the current flow never enters the tunnel.
Cryptokey routing makes destination and source prefixes part of peer selection. An outgoing packet is assigned to a peer based on the destination's allowed prefixes; a received packet is accepted against the sending peer's permitted source prefixes.[3] A valid handshake with peer A says nothing about a packet that the operating system routes elsewhere or that WireGuard assigns to peer B.
Choose one destination address and one protocol you are authorized to test. Prefer a service with a known expected response on the remote network, such as an internal gateway address or a controlled web endpoint. Record the exact time, destination, address family, source interface, and expected result.
Use a protocol-appropriate numeric-address probe with a known expected response. For HTTPS, preserve the original hostname, SNI, and certificate validation while directing the connection to the chosen IP with a documented client option. A name or certificate mismatch from replacing the URL hostname with an IP is not evidence of a routing failure; do not disable TLS verification.
Avoid beginning with a hostname. DNS introduces a second request, a resolver route, caching, and possibly several IPv4 and IPv6 results. Use a numeric destination first; test the name only after the packet path works.
Record the target peer's latest handshake time and TX/RX totals on both ends immediately before the probe. Then send a small, bounded request and record the same fields again. Do not continuously generate traffic because overlapping probes make deltas ambiguous.
Stop here if the latest handshake is not current for the intended peer. That is a handshake or identity problem, not the post-handshake state addressed here. The handshake-failure guide helps separate reachability from authentication.
Use direction labels from the test initiator's perspective. “Client TX” and “server RX” are the initiating and remote peers' accounted transfer bytes; these totals can include handshake and keepalive traffic. Exact byte totals may include protocol overhead, so seek timing and directional correlation rather than identical numbers.
For ordinary data in Linux v6.12, allowed-source validation precedes peer RX accounting; rejected source prefixes do not increment RX for that packet. Handshakes and keepalives also contribute to RX, so correlate the probe before inferring inner delivery.[4] Do not assume other implementations expose the same boundary.
| Fresh observation during one probe | First place to inspect | Why |
|---|---|---|
| Client TX does not rise | Client route, source policy, peer selection | The probe did not become transport data for this peer |
| Client TX rises; server RX does not | Outer path; receive validation if outer arrival is confirmed | The remote peer has not accounted for the probe |
| Server RX rises for ordinary probe data; no forwarded packet appears | Local input/forward choice, firewall, routing | Allowed-source validation passed in the cited Linux implementation; later delivery is unproven |
| Server forwards; no reply returns | Destination service, upstream filtering, NAT or return route | The request left the peer but no usable reply came back |
| Reply reaches server; client RX does not rise | Return peer selection, allowed source prefix, outer return path | The response did not become received client transport data |
| Client TX and RX both rise | DNS, MTU, TCP/TLS, or application | The encrypted path carried data in both directions |
One row is not a final diagnosis. It is the next evidence boundary. Confirm the same pattern twice with the same probe before changing configuration.
An optional independent control on Windows or Android is to disconnect your own peer, connect AethoVPN to a recommended location, and repeat one page load or short download in a fresh time window. Success establishes only that application task on the control path; it does not separate device faults from peer faults. The app shows connection states, not raw peer counters or packet captures, so read the matrix from your own peers after reconnecting them; the comparison does not diagnose or repair the original peer. If you choose to run it, start the 3-day free trial with your email.
Ask the operating system which route and source address it chooses for the exact destination. An interface can be active and a peer can have a recent handshake while a more specific main-table route, policy rule, local subnet, or another tunnel wins.
Then compare that destination with every peer's AllowedIPs. Overlapping entries or an unexpected narrow prefix can select a different peer. If you operate several peers, record the counters for all plausible peers instead of looking only at the one you expected.
Confirm that the application is not bound to another interface or address and that its network namespace sees the same routes you inspected. Containers, virtual machines, per-app VPN controls, and split-tunnel exclusions can have a different view from the host shell.
If this is an application-coverage question, full-tunnel VPN versus app proxy explains which traffic each approach takes over. Coverage differences do not change the evidence required for this WireGuard probe.
Make only one supported route or peer-selection correction. Retest the same probe and restore the original state if the selected interface or counter delta does not change.
A recent handshake and a lost transport packet can coexist because network state changed after the handshake. The remote endpoint may have roamed, a NAT mapping may have expired, a mobile network may have switched paths, or a firewall may treat later packets differently.
Compare the endpoint address currently recorded by each peer with the intended session, without publishing private addresses unnecessarily. If the roaming peer initiated the recent handshake, the other peer normally learns its most recent authenticated endpoint.[3] A stale observation from before a network handoff is not sufficient.
Use authorized packet or firewall counters on the outer interfaces to find the first missing direction. Do not expose private keys or disable the host firewall. If a new handshake immediately makes one data probe arrive, record that temporal relationship; it suggests path or state expiry, not a key mismatch.
The remote peer did not account for this probe in that window. Check outer-path evidence first. If the packet reached the outer interface, inspect receive authentication, replay checks, and allowed-source validation rather than declaring network loss.
After confirming RX belongs to this ordinary-data probe, allowed-source validation has passed in the cited Linux implementation. Once server RX rises, decide whether the inner destination is the server itself or beyond it. For a service on the server, verify local input firewall and whether that service listens on the WireGuard address, the expected port, and the expected address family. For a destination beyond the server, verify IP forwarding, the forward firewall, and the intended egress interface.
Next establish how replies return to the client prefix. A NAT design needs a correctly scoped translation rule and state. A routed design needs upstream knowledge of the WireGuard client network. If the destination is another WireGuard peer, the server needs an unambiguous peer mapping in each direction.
Do not add broad NAT automatically. It can make one probe work while hiding an incorrect routed topology. The connected-without-internet checklist covers full gateway, DNS, and dual-stack diagnosis once the counter pattern points there.
Fresh client TX and RX increases correlated with the probe show that encrypted transport data moved in both directions. If the user-visible action still fails, stop rotating keys or editing peer identities. Test the next layer.
First compare a numeric address with a hostname. Then compare a very small request with a larger response and repeat separately over IPv4 and IPv6. Small bidirectional traffic followed by a large-transfer stall is evidence for a WireGuard MTU problem, while a name-only failure points to resolver policy. A TCP reset, TLS alert, HTTP error, or application authorization response is also evidence that the tunnel carried the exchange far enough for a later layer to answer.
Application timeouts need correlation. Confirm that the TX/RX increase belongs to this request rather than a keepalive or background process. If possible, pause unrelated traffic or use a unique controlled destination and time window.
Write down the initial handshake time, four counter snapshots, route result, endpoint observations, and the first hop where evidence disappeared. Keep only the smallest correction that moved that boundary and is valid under the intended design.
Remove temporary routes, firewall rules, packet generators, and test services. If you changed a persistent peer mapping, verify the administrator's source of truth so a restart does not restore the fault. wg show all dump can expose actual interface private keys and peer preshared keys, not merely key status.[2] Never share raw output; use narrowly selected, sanitized fields instead.
If you suspect identity configuration despite a current handshake, identify which peer and which key generation the handshake actually used. The WireGuard key-mismatch guide explains a safe public-key comparison. Silence or a counter gap by itself cannot tell you which side is wrong.
An independent tunnel can provide an application comparison, but cannot isolate device versus peer faults. A successful request establishes that task on its own path; endpoint, port, protocol, routing, and device-policy differences remain. If both fail, inspect shared local conditions without treating a shared cause as proven.
Keep the comparison narrow. Run the control only while your own peer is disconnected so two tunnels do not compete for the default route, then reconnect your peer before taking new counter snapshots. A different server, port, and protocol also mean that a working control cannot confirm your own peer's endpoint or keys.
No. The peers can authenticate while the tested destination selects another route or peer, or while an inner source address is not permitted for the receiving peer.
They are cumulative and may include an earlier session, keepalives, or another flow. Snapshot both ends immediately before and after one controlled request and compare only the delta.
Yes. A path, endpoint, or NAT mapping can change after the last exchange. Correlate endpoint and outer-interface evidence with the exact failed probe.
It shows that the probe left the local peer but was not counted by the remote peer during that window. Confirm outer-path arrival first; if the packet did arrive, look at receive-side authentication, replay and allowed-source checks before calling it network loss.
First confirm RX belongs to this ordinary-data probe. In the cited Linux implementation, it has passed allowed-source validation; inspect local input versus forwarding, firewall policy, forwarding state, and egress routing next. RX alone does not prove the destination or application received it.
No. Bidirectional transport data is evidence that the current peer relationship works for that probe. Move upward to DNS, packet size, transport, or application behavior.
No. A keepalive maintains state but does not exercise your destination, response size, DNS lookup, or application protocol. Use a separate bounded request.
Disclaimer: This guide provides general technical troubleshooting information. Inspect only systems and traffic you are authorized to administer, minimize captured data, and never disclose private or preshared keys.
Sources:
Sources checked 9 October 2026.
Related Articles:
Sign up to experience all premium features at no cost.
*Available only to new users. Each user is limited to one trial.