Fix it now
This is a name resolution failure wearing a VPN costume. Before any tunnel exists the client has to turn the server’s name into an address using ordinary system DNS, and 868 is what you get when that step produced nothing. The VPN server may be perfectly healthy and your credentials perfectly valid.
nslookup vpn.example.com
nslookup vpn.example.com 1.1.1.1
ipconfig /flushdns
ipconfig /all
- Compare the two lookups. If the public resolver answers and yours does not, the fault is the client’s DNS configuration or something filtering lookups on that machine, not the VPN.
- If neither answers, read the name in the profile character by character. Open the connection’s properties from the Windows Settings VPN page, or from Network Connections with
ncpa.cpl, and check for a typo, a trailing space or an internal suffix appended to a public name. - As a test only, put the public IP address in the profile in place of the name and connect. Success confirms the diagnosis and nothing else.
- Fix the record properly, then put the name back. Certificate-based tunnels validate the name the client dialled against the server’s certificate, and an address will usually not match.
If the machine is behind a captive portal, complete the sign-in first. Nothing resolves until you do, and the VPN client reports it exactly like a broken DNS record.
If the connection now completes you can stop here. If it does not, the next section covers what the client is doing before the tunnel exists and where resolution usually breaks.
Why it happens
A VPN client is an ordinary network application until the tunnel is up. It uses the adapter’s configured DNS servers, the hosts file and the resolver cache exactly as a browser would. Anything that breaks name resolution on the machine breaks the VPN first, and it does so with a message about the remote access server rather than about DNS, which sends people to the wrong console.
The most common structural version of this is split-brain DNS. An internal zone holds a record for the VPN endpoint pointing at its private address, and the public zone either has no matching record or holds a different one. Inside the office the name resolves and the tunnel is unnecessary; outside, where the tunnel is needed, the name resolves to nothing. The endpoint needs a public record that works from a network you do not control, and the only honest test is to try it from one.
The codes that keep this one company point at neighbouring problems, and Microsoft publishes all three. 769 is the specified destination not being reachable, which usually means the adapter is disabled or has no address. 623 is not being able to find the phone book entry, which happens after profiles are recreated or deployed under a different name. 629 is the port being disconnected by the remote machine, which means resolution succeeded and something later in the sequence failed – a different article entirely.
The name in the profile is wrong or has changed
You have this one if nslookup fails on the same string from every network and every resolver.
- Compare the string in the profile against what the endpoint is published as, character by character.
- Watch for a trailing space, a character pasted from a document, or an internal suffix appended to a public name.
- Correct the profile, or recreate it, and connect again.
The client has no working DNS at all
You have this one if Every name fails, not only the VPN endpoint, and the adapter may show a limited or unidentified network.
- Run
ipconfig /alland check the adapter has an address and DNS servers. - Renew with
ipconfig /releaseandipconfig /renew, then flush withipconfig /flushdns. - Where a captive portal is in the way, sign in to it first.
The endpoint has an internal record only
You have this one if It resolves and connects from inside the network and fails from everywhere else.
- Publish a public record for the endpoint name pointing at the address the outside world can reach.
- Keep the internal record if internal clients need it, but make sure each answers with the address appropriate to where the client is.
- Test from a network outside your control before declaring it fixed.
A filtering product or a private resolver is answering badly
You have this one if A public resolver answers and the machine’s configured resolver does not, or returns nothing.
- Check whether a security product, parental filter or encrypted DNS setting is intercepting lookups on that machine.
- Where encrypted DNS is configured, confirm the chosen provider can see the record you need.
- Add an exception for the endpoint name where the product supports one, rather than turning filtering off.
A stale cache or hosts file entry is answering wrongly
You have this one if The name resolves to an address that is no longer correct, or resolution fails on one machine only.
- Flush the cache with
ipconfig /flushdnsand test again. - Inspect the hosts file under the Windows system directory, in drivers, etc, for an old entry naming the endpoint.
- Remove any stale entry. The hosts file overrides DNS entirely and will keep doing so after the record is fixed.
Editing the hosts file needs administrative rights, and an entry left there after a migration is one of the harder faults to spot because it affects one machine only.
Full reference
Sorting resolution failures from everything else
| What you see | Where the fault is |
|---|---|
868, and nslookup fails for the same name |
Name resolution on the client, or a missing public record |
| 868, but a public resolver answers correctly | The client’s own DNS servers, or a filtering product on the machine |
| Works inside the office, fails outside | An internal-only record for the endpoint |
| 769 | The adapter is disabled or has no usable address |
| 623 | The profile the client was asked to dial does not exist under that name |
| 629 | The far end was reached and then closed the port; resolution was fine |
What the three companion codes are published as
| Code | Published meaning |
|---|---|
| 629 | The port was disconnected by the remote machine |
| 769 | The specified destination is not reachable |
| 623 | Cannot find the phone book entry |
623 is the one worth understanding, because its cause is usually invisible. Windows keeps VPN connections in phone book files: one under the user’s application data for connections that belong to that user, and one for connections available to every user on the machine. A profile deployed by a management tool is written to one of them, so a connection somebody thought they deleted can persist, and a script dialling a connection by name can be dialling something that no longer exists under that exact name.
Testing name resolution properly
| Question | Command |
|---|---|
| Does my resolver answer for this name | nslookup vpn.example.com |
| Does anyone answer for this name | nslookup vpn.example.com 1.1.1.1 |
| What does this machine think its DNS servers are | ipconfig /all |
| Is something cached wrongly | ipconfig /flushdns then repeat the lookup |
| Is the record right for outside clients | Run the same lookup from a network you do not control |
Do all of this with the tunnel down. Once a tunnel is up the client may be using the tunnel’s DNS servers, which can resolve the endpoint name perfectly and tell you nothing about why the connection failed in the first place.
When the record resolves and the connection still fails
- Confirm the address it resolves to is the one you expect, not an old edge device that still answers.
- Check whether the profile carries a name with a trailing dot or an appended search suffix, which resolves differently from what you typed into nslookup.
- Check the certificate on the server if the tunnel is certificate-authenticated, because the name the client dialled has to match it.
- Look at whether the client is on a network that resolves the name and blocks the tunnel, which produces a different error entirely and is covered in the article on 809 and 800.
- Confirm no VPN client of its own is installed alongside the built-in one and intercepting the profile.
An IP address in the profile is a diagnostic, not a destination. It bypasses name validation on certificate-based tunnels, and it means every client needs editing the day the address changes – which is exactly the day nobody remembers where the addresses were hard-coded.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
868 |
Reported by the Windows VPN client when it cannot turn the name in the profile into an address. Microsoft publishes no meaning for the number, so treat it as a prompt to test resolution outside the tunnel | not published by the vendor |
629 |
The port was disconnected by the remote machine | Microsoft Learn |
769 |
The specified destination is not reachable | Microsoft Learn |
623 |
Cannot find the phone book entry | Microsoft Learn |
Confirm the fix worked
- Run
nslookupfor the endpoint name on the client with the tunnel down and confirm a correct answer. - Connect and confirm an internal address is issued.
- Test from a second external network, to prove the public record works generally rather than from one place.
- Confirm the profile uses the name rather than an address, so certificate validation continues to work.
- Flush the resolver cache and connect once more, so you know the success is not a cached answer.
Questions people ask about this
Does fixing this cost anything?
No. It is a DNS record, a profile setting or the client’s own network configuration. If a public record has to be created, that is a change at whoever hosts your domain rather than a purchase.
Can I just put the IP address in the profile permanently?
It is a useful test and a poor arrangement. Certificate-based tunnels validate the name the client dialled against the server’s certificate, and an address usually will not match. It also means every client needs updating when the address changes.
Why does it work on my phone but not on the laptop?
Different resolvers. The phone is probably using the carrier’s DNS while the laptop uses the local network’s, and one can see the record while the other cannot.
Where do VPN profiles actually live?
In phone book files: one under the user’s application data, and a second location for connections available to every user on the machine. Profiles deployed by a management tool are written there too, which is why an old profile can persist after somebody thinks they deleted it.
Is 868 always DNS?
It is always the client failing before the tunnel exists, at the point where it needs an address. That is nearly always name resolution, but the same symptom appears where the profile holds a name nobody ever published, so check the string as well as the resolver.
