Table of Contents
Wireguard Access
Summary
We use Wireguard for VPN access to the Sepia lab. Lab access requests are now automated.
How To
Summary
- Install Wireguard
- Generate a public/private keypair
- Provide the public key in the Onboarding Portal
- Once your access is approved, you'll be e-mailed your Wireguard IP
- Connect
Mac/Linux
1. For Mac, Install Wireguard from the App Store
For Linux,
sudo apt install wireguard-tools resolvconf # or sudo dnf install wireguard-tools
For Mac, install wireguard-tools using Homebrew
brew install wireguard-tools
3. Create a directory for a Wireguard keypair. E.g.,
mkdir ~/.wireguard
4. Generate a keypair
wg genkey | tee ~/.wireguard/private.key | wg pubkey > ~/.wireguard/public.key
Keep the Private Key secret!!
5. Provide the Wireguard public key (~/.wireguard/public.key contents) in the Onboarding Portal.
DO NOT ping any Lab Admins directly
6. Create ~/.wireguard/client.conf (note that the $PRIVATE_KEY must expand to
the actual text of your private key)
PRIVATE_KEY=$(cat ~/.wireguard/private.key) cat <<EOF > ~/.wireguard/client.conf [Interface] PrivateKey = $PRIVATE_KEY Address = X.X.X.X/32 DNS = 10.20.192.13, front.sepia.ceph.com, ipmi.sepia.ceph.com MTU = 1200 [Peer] PublicKey = kyEHy3ZBewI5RiK4/a0/UQn6O1kMt3h8V3u0OwsfUXc= AllowedIPs = 172.21.0.0/20, 172.21.32.0/20, 172.16.48.0/24, 10.20.192.0/20, 10.20.208.0/20, 172.16.50.0/23, 172.16.53.0/25, 172.16.55.0/26, 172.16.56.0/23, 172.16.59.0/25, 172.16.60.0/25, 172.21.64.0/24, 172.21.65.0/24, 172.21.66.0/24, 172.21.67.0/24 Endpoint = 192.86.31.5:1194 PersistentKeepalive = 25 EOF
6. Once you receive your private IP, replace X.X.X.X in client.conf with it.
7. Bring up the interface
On Mac OS, open the Wireguard GUI. Press Command+O and open ~/.wireguard/client.conf. Click Activate
On Ubuntu,
sudo mkdir -p /etc/wireguard sudo mv ~/.wireguard/client.conf /etc/wireguard/wg0.conf sudo chmod 600 /etc/wireguard/wg0.conf sudo wg-quick up wg0
Use wg show to show status.
Windows
1. Install Windows client from https://www.wireguard.com/install/
2. In the Wireguard app, select “Add tunnel/Add empty tunnel”. A public and private key will be automatically generated for you. Note them. KEEP THE PRIVATE KEY SECRET.
3. Submit your public key in the Onboarding Portal
4. Wait to receive your Wireguard IP
5. Add the configuration to the wireguard app. The first two lines ([Interface] and PrivateKey = <your private key>) will already be present. Be very careful to not change anything besides your Address. In particular, do not edit [Peer] PublicKey; that is the server's public key and does not change.
[Interface] PrivateKey = <your private key> Address = <address from communication with David/Dan>/32 DNS = 10.20.192.13, front.sepia.ceph.com, ipmi.sepia.ceph.com MTU = 1200 [Peer] PublicKey = kyEHy3ZBewI5RiK4/a0/UQn6O1kMt3h8V3u0OwsfUXc= AllowedIPs = 172.21.0.0/20, 172.21.32.0/20, 172.16.48.0/24, 10.20.192.0/20, 10.20.208.0/20, 172.16.50.0/23, 172.16.53.0/25, 172.16.55.0/26, 172.16.56.0/23, 172.16.59.0/25, 172.16.60.0/25, 172.21.64.0/24, 172.21.65.0/24, 172.21.66.0/24, 172.21.67.0/24 Endpoint = 192.86.31.5:1194 PersistentKeepalive = 25
6. Save the configuration, which should activate the wg0 tunnel. If all goes well, you will see “Latest Handshake” in the Peer section be a few seconds out of date
7. despite DNS = , the current wireguard client does not properly handle split-horizon DNS. You must add Powershell scripts to the configuration, and enable their execution with a Windows Registry setting.
Add the following after MTU = in the [Interface] section:
PostUp = powershell -ExecutionPolicy Bypass -File "C:\Wireguard\wg-up.ps1" -InterfaceAlias wg0 PostDown = powershell -ExecutionPolicy Bypass -File "C:\Wireguard\wg-down.ps1" -InterfaceAlias wg0
and add the scripts to C:\Wireguard:
# wg-up.ps1
param(
[string]$InterfaceAlias = "wg0",
[string[]]$Domains = @("front.sepia.ceph.com"),
[string]$DNSServer = "10.20.192.11"
)
# Ensure admin
if (-not ([Security.Principal.WindowsPrincipal] [Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole] "Administrator")) {
Write-Error "This script must be run as Administrator."
exit 1
}
# Verify the requested interface exists and is up
$Interface = Get-NetAdapter -Name $InterfaceAlias -ErrorAction Stop
if ($Interface.Status -ne "Up") {
Write-Error "WireGuard interface /$InterfaceAlias' is not up."
exit 1
}
Write-Output "Using WireGuard interface: $InterfaceAlias"
# Add NRPT rules (idempotent)
foreach ($domain in $Domains) {
if (-not (Get-DnsClientNrptRule | Where-Object { $_.Namespace -eq $domain })) {
Write-Output "Adding NRPT rule for $domain -> $DNSServer"
Add-DnsClientNrptRule -Namespace $domain -NameServers $DNSServer
}
}
# Set connection-specific DNS suffix (short names)
Set-DnsClient -InterfaceAlias $InterfaceAlias -ConnectionSpecificSuffix $Domains[0]
Write-Output "Set connection-specific suffix: $($Domains[0])"
# wg-down.ps1
param(
[string]$InterfaceAlias = "wg0",
[string[]]$Domains = @("sepia.ceph.com")
)
# Ensure admin
if (-not ([Security.Principal.WindowsPrincipal] [Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole] "Administrator")) {
Write-Error "This script must be run as Administrator."
exit 1
}
Write-Output "Using WireGuard interface: $InterfaceAlias"
# Remove NRPT rules safely
foreach ($domain in $Domains) {
Get-DnsClientNrptRule |
Where-Object { $_.Namespace -eq $domain } |
Remove-DnsClientNrptRule -Force -ErrorAction SilentlyContinue
Write-Output "Removed NRPT rule for $domain"
}
# Clear connection-specific suffix
Set-DnsClient -InterfaceAlias $InterfaceAlias -ConnectionSpecificSuffix ''
Write-Output "Cleared connection-specific suffix"
Add to the registry:
HKEY_LOCAL_MACHINE\Software\WireGuard\DangerousScriptExecution
set it to 1. You can do this with a Command prompt running as administrator:
reg add HKLM\Software\WireGuard /v DangerousScriptExecution /t REG_DWORD /d 1 /f
DNS Configuration using dnsmasq on Linux
If you're using systemd-resolved, stop. It is incapable of handling split DNS. I would not use resolvconf.
The way I (dmick) chose on Fedora 41 was to disable systemd-resolved and set up a dnsmasq instance behind /etc/resolv.conf. Here's what I did:
1) sudo systemctl stop systemd-resolved; sudo systemctl disable systemd-resolved 2) add this to /etc/systemd/networkd.conf and restart networkd (because I'm running networkd to configure networks):
[Network] DynamicUser=no ManageResolver=false
3) install dnsmasq if necessary, and configure it. I chose to put my configuration in /etc/dnsmasq.d/dnsmasq.conf (because my system runs dnsmasq with a conf search path of /etc/dnsmasq.d):
# add domain to shortnames in /etc/hosts; may be helpful expand-hosts # don't use /etc/resolv.conf for servers configured here no-resolv # for debugging, enable log-queries # log-queries # set the server that should handle these three domains server=/sepia.ceph.com/10.20.192.11 server=/front.sepia.ceph.com/10.20.192.11 server=/ipmi.sepia.ceph.com/10.20.192.11 # set the upstream servers for anything else server=1.1.1.1 server=9.9.9.9
4) remove the existing /etc/resolv.conf and replace with this:
nameserver 127.0.0.1 search front.sepia.ceph.com ipmi.sepia.ceph.com sepia.ceph.com options ndots:2
5) systemctl restart dnsmasq
Now lookups of shortname or shortname.front or shortname.ipmi should work. Note that dig does not respect the search domains in /etc/resolv.conf by default; you must use dig +search <domain>
Split DNS on systemd-resolved distros (Fedora, recent Ubuntu, etc.)
If all web browsing / external DNS stops working while connected to the Sepia VPN, but works again after sudo resolvectl revert sepia, your distro is using systemd-resolved and wg-quick has set the lab DNS server as the default resolver for all queries — not just Sepia ones.
Do not use resolvectl revert as a workaround; it also breaks resolution of lab hostnames. Configure split DNS instead so only Sepia domains are sent to the lab resolver:
- Edit
/etc/wireguard/sepia.conf - Remove (or comment out) the
DNS =line - Add the following lines under
[Interface]:
PostUp = resolvectl dns %i 10.20.192.13 PostUp = resolvectl domain %i ~sepia.ceph.com ~front.sepia.ceph.com ~ipmi.sepia.ceph.com PostUp = resolvectl default-route %i false
Then restart the tunnel:
sudo wg-quick down sepia && sudo wg-quick up sepia
Verify:
resolvectl status sepia
You should see Default Route: no and the three sepia domains listed. Lab hostnames (e.g. smithi001.front.sepia.ceph.com) will resolve via the VPN; everything else uses your normal DNS.
Alternative: NetworkManager
On Fedora you can instead import the tunnel into NetworkManager, which handles split DNS natively:
sudo nmcli connection import type wireguard file /etc/wireguard/sepia.conf sudo nmcli connection modify sepia ipv4.dns-search "~sepia.ceph.com;front.sepia.ceph.com;ipmi.sepia.ceph.com" ipv4.dns-priority 50 sudo nmcli connection up sepia
Only use one method — don't run wg-quick and the NetworkManager connection at the same time.

