What's inside the Vates VMS hardening guide
The Vates VMS Hardening Guide, written by the Vates VMS Security Team, collects security recommendations for
deploying and running Vates VMS (the Vates Virtualization and Management Stack, made of XCP-ng and Xen
Orchestra). This revision, dated November 10, 2026, targets XCP-ng 8.3 LTS and Xen Orchestra 6 (the "stable" and
"latest" release channels of XOA, or installations from the sources tracking the master branch): these are the
versions that receive security updates. XCP-ng 8.2 LTS reached its end of life on September 16, 2025, so if you
still run it, upgrading to 8.3 LTS is the first hardening step. The guide is written for administrators, systems
architects and security specialists who already know XCP-ng pools, networking and the role of the pool master.
It covers five layers of the stack (network and storage, the virtualization platform, virtual machines,
orchestration, and backups) and recommends starting with a risk assessment, for example with the EBIOS Risk
Manager framework. It draws on guidance from NIST, ENISA, ANSSI and CISA.
Network isolation
The Vates VMS hardening guide treats physical network isolation as the basis of a defense-in-depth strategy: if
one segment is compromised, the breach does not automatically reach the rest of the infrastructure. XCP-ng can
carry management, storage and guest traffic on a single interface, but the guide recommends separating them.
-
Management network:
it carries host-to-host communication, live migration, VM import and export, and the Xen Orchestra connection.
It must never be reachable from unsecured or public networks, and never from the Internet: remote
administrators reach it through a VPN or a Zero Trust access gateway.
-
Storage network:
a dedicated physical network (or VLANs, if you accept weaker isolation), bonded for redundancy, on its own
subnet that is not routable to other networks.
-
Backup network:
it is the management network by default. You can give Xen Orchestra a second virtual interface on a dedicated
backup network so that a compromised NAS cannot reach the hosts' management network. Backups sent to S3 should
travel over HTTPS or a tunnel.
-
VM network:
VMs are treated as untrusted. Use a dedicated NIC with no IP address in Dom0, and ideally give VMs separate
networks for their own administration and for public services.
The guide also covers private networks with GRE/VXLAN tunnels managed by the Xen Orchestra SDN controller (with
IPsec encryption when they cross links you do not control), port locking (the
default-locking-mode
network setting) to block ARP and IP spoofing, and traffic rules. These rules filter VM traffic in the virtual
switch of each host, on a single virtual interface or a whole network, including VLANs and bonds, and are
managed from XO 6, the REST API or the
xe
command. Further advice: give every pool member the same number of NICs, use readable network names, and apply
consistent cabling conventions.
Never expose Xen Orchestra to the Internet
This is the main change in this revision. Earlier versions accepted an Internet-facing Xen Orchestra protected
by strong authentication. The guide no longer does, because AI-assisted attacks have cut the time between a fix
and a working exploit to hours, and exposed interfaces are found and attacked within hours of going online.
Multi-factor authentication protects the login form, not the code that runs before it. Xen Orchestra, XAPI, XO
Lite, XO proxies, server BMCs and backup repositories must never be reachable from the Internet, in line with
CISA Binding Operational Directive 23-02 and the ANSSI recommendations on secure administration.
-
Recommended access:
a dedicated administration network reached from dedicated workstations, through a VPN with phishing-resistant
MFA (WireGuard or IPsec rather than legacy SSL VPNs) or a Zero Trust Network Access gateway, plus a bastion
for privileged sessions and site-to-site tunnels between sites.
-
Not a substitute:
port forwarding, a reverse proxy that only forwards traffic, IP allowlists, geoblocking, port knocking or a
hard-to-guess host name.
-
The gateway:
it becomes your most exposed asset. Place it in a DMZ, patch it within days, and never expose its own
management interface.
-
Verify from the outside:
scan your public IP ranges regularly. If Xen Orchestra or a host was exposed, remove the exposure, treat the
system as compromised, review the audit log and rotate every credential it holds.
Storage protection
The Vates VMS hardening guide recommends dedicated storage NICs on a different subnet from management. It also
recommends naming Storage Repositories by role, disk type and location (for example Prod_SSD_Chicago) and making
sure that every site holding storage is physically as secure as the first one. For local storage, it recommends
RAID (hardware or software) with regular health monitoring, plus off-host backups. Advice for each storage
protocol:
-
NFS:
restrict exports to specific IP addresses, because virtual disk files (VHD or QCOW2) are not encrypted by
default, and keep the storage network isolated. The default AUTH_SYS flavor trusts the client's IP address and
does not authenticate hosts, so these two controls are what actually prevent unauthorized access.
-
SMB:
separate traffic with VLANs, use username and password host authentication, and never let a production pool
share a storage network with a test pool.
-
iSCSI:
separate traffic with VLANs or a dedicated physical network, use CHAP for both target and host, and use
multipath across different switches.
-
Fibre Channel:
use LUN masking per HBA and SAN zoning.
Hardening XCP-ng hosts
Before installing XCP-ng, the guide recommends applying vendor firmware, BIOS/UEFI and CPU microcode updates,
and checking the official requirements and hardware compatibility list. Download the most recent ISO from
xcp-ng.org, check its GPG signature and SHA256 fingerprint, and select "Verify Installation Source" so that the
installer checks the GPG signatures of the packages.
Security updates are published as soon as they are ready, and since 2026 two or three security batches in a
single month are common, partly because of AI-assisted security research. They are never gated behind a paid
support contract. Each one is documented in a Vates Security Advisory (VSA) with a Vates severity rating:
subscribe to the VSA RSS feed to prioritize the updates that matter. Apply them pool by pool, master first, for
example with Rolling Pool Update. Exchanges inside a pool must stay on an isolated network, and sensitive
workloads can be placed in separate pools with their own network and storage. If a pool join fails on a
certificate error, synchronize the pool certificates instead of weakening TLS verification. Unless you need it,
do not join pools to Active Directory or grant XAPI roles: manage users in Xen Orchestra.
For the control domain (Dom0), the guide's rule is to make no changes unless Vates recommends them, and to add
no services or local users: keeping Dom0 minimal is what kept the 2026 Linux kernel privilege escalations out of
reach. Use strong passwords, change the root password for the whole pool from the master, and rotate the pool
secret and root passwords when someone with access leaves. Other recommendations: synchronize time with NTP (TLS
certificates and tokens depend on it), send logs to a remote syslog server that XCP-ng administrators cannot
access, and turn off plain HTTP with
xe host-param-set https-only=true
on each host. You can disable SSH from xsconsole (Remote Service Configuration), because Xen Orchestra does not
need SSH to manage hosts. If you keep it, use ED25519 keys and allow it only from the administration network.
The Xen Orchestra Audit Log uses a hash chain to make the record of user actions tamper-evident, and records the
client IP address of each action.
Physical security covers the supply chain (trusted suppliers, certified refurbishers), locked and monitored
rooms, BMCs (IPMI, iDRAC, iLO) that are never reachable from the Internet and sit on an isolated network with
encryption and IP filtering, and decommissioning hosts with NIST-grade data sanitization and an audit trail.
Protecting Xen Orchestra
The guide describes how the XOA web deployment tool works: it sends no data back to Vates. One caveat is that
the XOA template is downloaded over HTTP because of XAPI limitations, so if that is a concern you can download
it manually and check its MD5 checksum. XOA only needs outbound HTTPS to xen-orchestra.com and nodejs.org for
its updates. Replace the self-signed certificates with trusted ones on all three channels: the web interface,
XAPI and stunnel, and the SDN controller. Let's Encrypt with the DNS-01 challenge gives automatic renewals
without exposing Xen Orchestra, and SDN controller certificates must be regenerated for OpenSSL 3.
Xen Orchestra ships a release on the last working day of each month. Security fixes reach the "latest" channel
first and "stable" with the next release, and serious advisories provide a mitigation for "stable" appliances in
the meantime. Keep the automatic safety snapshot taken before each update, so a failed update can be rolled
back. XOA runs on Debian with unattended security updates, but a kernel fix only takes effect after a reboot.
Moving to a new major Debian version means redeploying XOA and importing its exported configuration, not running
dist-upgrade. Further recommendations:
-
Encrypt the credential database at rest (AES-256-GCM, since Xen Orchestra 6.5), and keep Redis listening on
the local interface only.
-
Use strong passwords plus two-factor authentication, either built in or through SAML or OIDC (Keycloak, Okta,
Auth0), ideally with hardware keys such as YubiKeys. Keep the LDAP and SAML plugins up to date and correctly
configured.
-
New users start with no permissions. Xen Orchestra 6 adds RBAC for the REST API and XO 6: deny by default,
with roles such as Storage administrator or Network administrator to delegate an area without global rights.
Keep superadmin rights to a few people, and remove accounts as soon as they are no longer needed.
-
Connect XO proxies to the main site through a site-to-site tunnel (IPsec or WireGuard): neither the hosts nor
the proxies of a remote site should be exposed to the Internet.
-
For the REST API, prefer tokens to passwords, tied to a dedicated user with limited roles, with a capped
lifetime (maxTokenValidity), sent over HTTPS and never placed in URLs or browser storage.
-
Run the read-only MCP server for AI assistants inside the administration network, with a restricted token, and
check where a cloud-hosted assistant sends your infrastructure data.
-
Review integrations like user accounts: limit the OpenMetrics port to your Prometheus servers, and keep the
credentials of Terraform, Ansible or Kubernetes in a secret manager.
Virtual machines
The guide recommends securing VMs like physical servers: keep the attack surface minimal, start from hardened
templates, and automate deployment with Ansible, Packer or Terraform. Patch VMs immediately when they come back
after a long period offline. Create VMs in UEFI mode with Secure Boot whenever the guest supports it: XCP-ng 8.3
now ships the required certificates, including Microsoft's 2023 ones. Add a vTPM where needed (Windows 11,
BitLocker). Use the current XCP-ng Windows guest tools, signed by Microsoft, never the legacy 8.2 drivers, and
always create VMs from the template matching their operating system. Reserve PCI, GPU and USB passthrough to
trusted VMs, ideally on dedicated hosts.
Xen Orchestra names, descriptions and scoped, color-coded tags (for example "prod=database") reduce operational
errors, and the same tags drive smart backups and load-balancer anti-affinity rules. Size VMs to withstand
denial-of-service conditions without starving other VMs. Since the XO Load Balancer and High Availability can
move VMs automatically, make every host in a pool meet the same security criteria, or exclude certain hosts from
migrations. Each migration decided by the load balancer is logged as a task with its reason.
Management and privileges
General practices in the guide: HTTPS only, no direct exposure of the XCP-ng API or XO Lite (access only from
the administration network), up-to-date third-party tools that talk to XAPI, prompt patching guided by the VSA
feed, network segmentation and firewalls, auditing, and backups that are tested. Admin rights follow least
privilege, with unique accounts, MFA, revocation tied to HR workflows, and ongoing training. The guide also
advises keeping things simple: enable only the features you need. For example, a misconfigured High Availability
setup can spread a VM that crashes on boot from host to host.
Securing backups
The backup chapter defines the backup types (full, incremental, replication, mirror, metadata, smart and
immutable backups), points to NIST SP 800-209, and recommends the 3-2-1 rule or its 3-2-1-1-0 variant (one
immutable or air-gapped copy, zero errors on restore tests). It recommends an isolated backup network, MFA on
backup access, encryption with keys stored apart from the data, and keeping NAS firmware up to date. Untrusted
or third-party repositories must always be encrypted. Xen Orchestra configuration backups contain your
credentials: protect them with a passphrase.
Ransomware operators try to destroy backups before encrypting production, so the guide recommends immutable
backups: S3 Object Lock, or the Xen Orchestra immutable backups service on your own file server. The immutable
repository must be administered separately, so that a compromised Xen Orchestra cannot disable immutability. The
Auto Restore Check feature (the "Health check" option in a backup job) restores each VM, waits up to 10 minutes
for its guest tools to load, then deletes it. Complement it with regular manual restore tests of critical VMs.
Notifications by email, Slack or Teams, and monitoring through Nagios, Prometheus or Centreon, make sure failed
backups are noticed.
Security processes
The appendices describe how security issues are handled in Vates VMS. They reproduce the Xen Project security
process (usually one working week from report to advisory, then two working weeks of embargo before public
disclosure). XCP-ng has been on the Xen pre-disclosure list since 2018, and Vates engineers authored more than
10% of the commits of Xen 4.22. Fixes are published on the XCP-ng mirrors, in Xen Orchestra and in posts tagged
"security" on the blog. Xen Orchestra gets security support only on the "stable" and "latest" channels and on
sources tracking the master branch. Since September 2025, every advisory is published as a Vates Security
Advisory (VSA), with the upstream CVSS score and a Vates severity rating that reflects the real impact on
supported deployments. Vulnerabilities are reported to security@xcp-ng.org or security@vates.tech.
Key takeaways
-
Never expose Xen Orchestra, the hosts, XO proxies or BMCs to the Internet: go through a VPN or a Zero Trust
gateway.
- Physically separate management, storage, backup and VM traffic, and treat every VM as untrusted.
-
Run XCP-ng 8.3 LTS, follow the VSA feed, check the integrity of the installation media, keep Dom0 unmodified,
disable HTTP and SSH, and use remote syslog.
-
Protect Xen Orchestra with trusted certificates, MFA, least-privilege RBAC, an encrypted credential database
and XO proxies behind site-to-site tunnels.
- Use Secure Boot, vTPM and up-to-date guest tools in your VMs.
-
Isolate, encrypt and make backups immutable, and check them with Auto Restore Check and manual restore tests.
Download the guide (PDF)