Repository navigation
Releases
v1.2.2
A small release: feedback in 30 seconds, no account needed.
New
- Feedback without a GitHub account. The 💬 button at the bottom of the sidebar opens a short form: pick 😀 😐 😞 or 💡 and add a sentence if you like. Your VirtoMir, OS, libvirt and QEMU versions are filled in for you; host names, addresses and serial numbers are never sent. The demo's "Idea / Report a bug" link opens the same form.
- With a GitHub account, the same questions are available as an issue form ("Quick feedback").
Documentation
- The README translations are now generated from one template (
scripts/readme-i18n), so all 15 languages stay in step with the English page. - The AUR PKGBUILD in this repository names the release checksums.
Everything from v1.2.1 (security and data-safety fixes) is included.
VirtoMir v1.2.2-20-g5216c0b
Build of main at 5216c0b. Hosts on the main update channel install it; it is replaced by every push to main.
v1.2.1
VirtoMir 1.2.1 is a hardening release: no new features, but fixes found by testing on real hardware (Fedora 44 and Ubuntu 24.04, single hosts and a two-host cluster). Upgrading is recommended, especially for clusters.
Security
- Cluster join: a join request is now checked against the invite before the host connects anywhere. Before, anyone who could reach port 8420 could make a host open HTTPS connections to addresses of their choice.
- Off-site copies: a password inside the repository address (
rest:https://user:password@host/) was shown unmasked to every role that can read the off-site settings. It is now masked like the other secrets.
Fixes
- PCI passthrough on Debian and Ubuntu: turning the IOMMU on reported success, but on hosts with settings in
/etc/default/grub.d(all cloud images) the kernel came back without it. VirtoMir now writes its own file there and checks the boot entry before it asks for the reboot. - SELinux (Fedora, RHEL): storage pool folders and backup locations outside the usual places are now labelled for VM images by VirtoMir itself. Before, every disk operation there raised an SELinux alert (Fedora) or failed (enforcing hosts).
- Restore: a backup restored into a pool without room is refused before anything is written. The check now reads the pool's current free space, not libvirt's cached value.
Refused before anything changes
- More vCPUs than the host has CPU threads, on create and in the settings form.
- More memory than the host has in the settings form (only create checked it).
- A live migration to a host without enough free memory.
- A network adapter on a bridge or virtual network the host does not have (the machine would not start again).
Smaller changes
- Raising the vCPU or memory maximum of a running machine no longer shows libvirt's raw error; the change applies at the next start.
- The Ceph, off-site and BGP pages name the cluster members that still lack a package the open host has.
- A new README with a tour of the UI, current screenshots and a menu of good first issues, translated into 15 languages.
VirtoMir v1.2.0
VirtoMir 1.2 takes the jobs you used to do by hand on the host and does them itself. You no longer have to type storage addresses, install packages, edit boot settings or copy SSH keys. It also closes a long list of security and data-safety gaps.
Highlights
- Network storage, found and connected for you. VirtoMir scans the network for NFS, SMB and iSCSI shares, and finds Fibre Channel ports, SAN LUNs and multipath devices. One click makes a pool from any of them. iSCSI pools support CHAP.
- Pools of a chosen size from free space. Make a pool from unused space on a disk; removing it gives the space back.
- Datastore browser for every file-based pool: look around, upload and download, and see which disks are in use.
- Bridge machines to Wi-Fi, like VirtualBox and VMware Workstation do. Machines get an address from your router, not from NAT. Any host interface can now carry a bridged network.
- Nothing to install by hand. When a feature needs a tool the host lacks (restic, FRR, dnsmasq + iPXE, cephadm + podman), an Install it button sets it up with apt or dnf. In a cluster, one click installs it on every host. Progress shows in Tasks.
- One-click host setup. VirtoMir now handles these steps itself:
- PCI passthrough: it turns on the IOMMU on one host or on all of them, keeping a backup of the boot settings. Hosts whose kernel already has it on are left alone.
- High availability: it loads a watchdog on every host when you enable HA.
- ZFS replication: it sets up the SSH key on the target. For a host outside the cluster it asks for that host's root password once and does not keep it.
- Templates from running machines. One click cleans up a running machine, shuts it down and turns it into a template.
- Machine timeline. One list on the machine page shows everything that happened to a machine: power events, changes, snapshots, backups and alerts.
- Update channels. Follow
main(a signed build of every change) or stable releases only. Automatic updates can be switched off, and you can check for updates by hand. - Easier DRS rules. You pick machines by name instead of typing UUIDs. Applying a recommendation really moves the machine, and the move shows in Tasks.
- Better on Fedora and RHEL. VirtoMir now ships an SELinux policy module, including rules for libvirt's storage daemon. It works with modular libvirt daemons, and network installs find iPXE where Fedora installs it.
- Wide screens are used, and there is an Idea / Report a bug link at the bottom of the sidebar.
Security
- Every response now carries security headers, and the UI runs under a Content Security Policy.
- A session follows its account's role: a demoted or removed user loses their rights at once.
- Failed sign-ins are counted per IPv6 network, so rotating addresses does not get around the lockout. Old sessions are cleared.
- A single sign-on can only be finished by the browser that started it.
- The updater checks that the signed checksum belongs to the exact package being offered.
- A storage connect request can never smuggle in a path as the pool name.
- The memory VirtoMir uses to answer repeated writes is now limited.
- ZFS replication: a target whose SSH identity has changed is refused. The root password is used for one sign-in and is never stored or logged.
Data safety
- A running machine's disk can no longer be made smaller. This is checked in the daemon too, not only in the UI.
- A machine can no longer ask for more memory than the host has. This is also checked in the daemon.
- An OVA that does not fit is refused before it is unpacked.
- Settings survive a power loss and a binary update, because every settings file is written atomically.
- An unreadable settings file is kept, never overwritten. This covers HA, pools, placement, replication, VMware and off-site copies, and the feature stays usable.
- A backup counts as finished only once it is on disk.
- Configuration history can restore an earlier version after a rename, says which disks are gone, and keeps versions in order when the clock goes back.
- These storage cases are now handled safely:
- A carved pool that is not mounted never has its space given back.
- The datastore browser keeps disks that are reached through links.
- The base image of a stopped machine's overlay counts as in use.
- Removing a pool also removes its CHAP password or Ceph key.
- A write the browser loses to a network change is not done twice.
- A stopping daemon lets cancelled jobs clean up, and a stop is not reported as a failure.
Fixes
- An empty list from a host no longer breaks a page. This was fixed in about 50 places, in the daemon and in the UI.
- Errors read as words, never as a raw JavaScript error. "No host answered" is no longer shown as a failure.
- The DRS card uses the same live numbers as the Host CPU tile. The Machines header counts follow the node filter.
- "Found on your network" no longer fails when a scan finds nothing. iSCSI, SCSI and multipath pools are never shown as "100% full".
- A disk on NFS is deleted even when libvirt refuses.
/srv/virtomirgets its SELinux label before a share is mounted. - No machine or bridged network is created on a bridge the host does not have. New subnets are fresh, and Wi-Fi cards are not offered as bridge ports.
- HA and DRS stay out of the way on a single host.
- A host with a hardware watchdog does not also get a software one.
- Wi-Fi bridging learns addresses only from its own machines, and never the gateway's.
- In a VMware migration wave, only one worker lists vCenter.
- "Last auto-upgrade 1/1/1" is no longer shown before the first upgrade.
- The online demo answers like a real host: saved settings stay, and the pools and networks you create appear.
Upgrading
Click Update in Settings → Updates. You can also upgrade from the package repositories (apt upgrade / dnf upgrade) or run install.sh from the new tarball. Settings, certificates and metric history are kept.
Thanks
Thanks to Mega Bilgisayar for supporting VirtoMir.
Full Changelog: v1.1.1...v1.2.0
VirtoMir v1.1.1
Full Changelog: v1.1.0...v1.1.1
VirtoMir v1.1.0
VirtoMir v1.0.0
VirtoMir 1.0: the feature set of the on-prem cloud is complete: Kubernetes clusters, hyper-converged Ceph, software-defined networking, disaster recovery and new hosts installed over the network. From here on the work goes into hardening what exists. The newest features were tested live on a nested two-host cluster, not yet on real hardware, so test reports help the most (#1).
New
- Kubernetes clusters. Create clusters on VirtoMir machines with kubeadm through cloud-init: one or three control planes (kube-vip for the shared address), flannel or calico, kubeconfig download, scaling and minor upgrades node by node. VirtoMir makes the cluster's certificate authorities itself. A CSI driver (integrations/virtomir-csi) gives pods volumes from VirtoMir storage pools as hot-plugged virtio disks.
- Bare-metal onboarding. New hosts boot from the network (iPXE, Debian 13 preseed), install VirtoMir from the apt repository and join the cluster on their first boot with the new
virtomird join. BMCs power and boot them through Redfish or IPMI. - Hyper-converged Ceph. Bootstrap Ceph with cephadm, add empty disks as OSDs, create RBD pools and storage policies, and watch health and the OSD tree in the Ceph deck.
- Storage efficiency. Compression and dedup per pool (ZFS, Ceph), compacting stopped disks, storage tiers with disk moves between them, and NVMe/TCP targets as pools. A machine's disk can move to another pool while it runs.
- Software-defined networking on OVN. Per-project VPCs with subnets and DHCP, logical routers, external gateways with SNAT and floating IPs, security groups, load balancers with health checks and IPsec between hosts.
- Physical network. SR-IOV networks, BGP (and EVPN) through FRR, IP pools that hand out fixed addresses to new machines, and dynamic DNS (RFC 2136 with TSIG).
- Disaster recovery. Protection groups, recovery plans with boot groups, test failover, real failover and failback with reverse ZFS replication, and RPO tracking.
- Rolling cluster upgrades. Pre-checks, one host after the other with pause, resume and abort, OS package upgrades and reboots, and a banner while versions are mixed.
- GPU pools. GPU inventory with MIG profiles and vGPU types, GPU pools with per-project quotas, and placement, moves and DRS that respect machines' GPUs.
- Insights. Capacity forecasts, right-sizing suggestions and anomaly alerts.
- Machines. Incremental backups with libvirt checkpoints, a virtio-win drivers CD for new Windows machines, what the guest agent reports and setting a password, traffic limits per network adapter, CPU weight and limit, and nested virtualization.
- Automation. A Packer builder (integrations/packer-plugin-virtomir) that makes templates from cloud images, a Kubernetes operator (integrations/virtomir-operator) for machines and projects, and Terraform data sources for machines, networks, pools and projects.
Fixed
- Security: updates and container images are only accepted with a valid OpenPGP signature; a checksum alone is no longer trusted (#36).
- DRS settings, placement rules and VXLAN overlays now reach cluster members that were down or join later (#37).
- Snapshots work on raw, LVM, ZFS and Ceph disks.
- An IP pool can no longer overlap its network's DHCP range; iSCSI LUNs and disks in use are no longer offered as Ceph OSDs.
Good to know
- Kubernetes, the CSI driver, bare-metal onboarding, Ceph, SR-IOV and BGP were tested on nested KVM only. Not tested yet: real BMCs, proxy-DHCP for network boot, and Kubernetes nodes spread over several hosts.
- Ceph needs cephadm and podman on the host, SDN needs OVN, BGP needs FRR.
- In a cluster, update every host to 1.0.0: the new settings are kept in sync between hosts and older members do not know them.
Full Changelog: v0.9.0...v1.0.0
v0.9.0
The first step from a virtualization manager towards an on-prem cloud: projects for teams, a self-service catalog, migration from VMware, placement rules, a content library and compliance checks. Every feature below was tested live on a two-host cluster before it was merged.
New
- Projects (multi-tenancy). Group machines, storage pools and networks per team, with members (users or groups) as viewer, operator or admin, and limits on machines, vCPUs, memory and disk. Project roles apply to the project's machines; people without a global role can sign in as member and see only their projects. Projects are the same on every host of a cluster.
- Self-service catalog. Admins publish blueprints (cloud image or template, size presets, allowed networks and projects). Project members order them; orders can need approval by a project admin. Sign-in credentials are asked for at order time. Optional leases with reminders, extension and automatic expiry, and showback: monthly cost per project from your own prices (HTML, CSV, JSON).
- Import from VMware. Connect to vCenter or ESXi (TLS fingerprint pinned, password never logged), browse the inventory and convert machines with virt-v2v as background jobs, with pre-flight checks, network mapping and waves of several machines in parallel with retry.
- Placement rules. Machine-to-machine affinity and anti-affinity, machine-to-host rules (must / should, run on / never run on) and host labels for racks and zones. Create, manual moves, DRS and HA restarts follow them; violations are shown.
- Content library. Templates, ISOs and cloud images with versions and SHA-256, copied to the other hosts of a cluster over the existing peer connection (resumable, checksum-verified), and machines deployed from the library on any host.
- Compliance. Hardening checks for virtualization hosts (a CIS/STIG subset: SSH, sVirt with AppArmor/SELinux, firewall, kernel settings, updates, time sync …) with reports and schedules; a tamper-evident audit log (hash chain) with verify and export; FIPS mode detection.
- Automation. New Ansible modules for networks, storage pools, projects, placement rules and snapshots (with check mode), and Terraform resources for projects and placement rules.
- Updates from the web UI for every install. Besides the .deb, hosts installed from the .rpm (dnf) or the tarball / install.sh, or running a binary anywhere, can now update themselves: the release file is checked against its SHA-256, swapped in place and the daemon restarted.
Fixed
- The power menu and the "more" menu of a machine no longer open underneath the cards below them.
- The updates card always shows the install button; when a host cannot install it explains why instead of hiding the button.
Good to know
- Import from VMware was tested against the VMware simulator (connection, fingerprint pinning, inventory, pre-flight and a real virt-v2v run); a full disk conversion from a real ESXi host has not been tested yet. Feedback welcome.
- FIPS enforcement of the daemon's TLS settings is experimental.
- In a cluster, update every host to 0.9.0: projects, placement rules and the content library are kept in sync between hosts and older members do not know them.
Full Changelog: v0.8.1...v0.9.0
VirtoMir v0.8.1
A safety release for high availability. Please upgrade every host of a cluster that uses HA.
Fixed
- A restarted daemon could stop protected machines at once when the other members were slow to answer (for example when all hosts are upgraded together). Votes now go missing only after the failover timeout.
- A host coming back could start a machine that already runs elsewhere. After its machines were restarted on another member, a returning host resumed them from its own outdated records. It now waits for the other members and resumes only machines it still owns.
- A split between two members could run a machine twice. With a TCP witness, both sides of a split each count the witness and keep quorum, so the other side's watchdog never resets it. HA no longer treats the watchdog as proof in that case.
Changed behaviour
In a two-member cluster with a witness, automatic failover now needs a fence agent (IPMI / Redfish) for each host, or an admin confirming on the HA page that the failed host is powered off. Clusters of three or more members are unchanged: the watchdog still counts as proof there, because a minority cannot keep quorum.
Tested on two hosts: a host powered off, the hosts cut off from each other with the witness reachable from both, and a host coming back after a failover.
Full Changelog: v0.8.0...v0.8.1
VirtoMir v0.8.0
VirtoMir 0.8.0 makes the features introduced in 0.7.0 work as described, and fixes a few serious bugs found while testing them on a two-host cluster.
Coming from 0.7.0 or 0.7.1? Backups of running machines failed there (manual and scheduled). 0.7.2 fixed that; if you skipped it, check that your recent backups exist after upgrading.
Now working for real
- System containers (LXC) — Debian 13, Ubuntu 24.04 or Alpine 3.22 from images.linuxcontainers.org, checked against SHA-256, unprivileged (container root is an unprivileged user on the host). Console sign-in, start, shut down, reboot, delete, live CPU/memory. Needs
libvirt-daemon-driver-lxcon Debian/Ubuntu. - VXLAN overlays — one layer-2 network across hosts over UDP 4789. Created cluster-wide, every member gets the others as peers and one host is the gateway (NAT, DHCP). VirtoMir keeps the tunnels in place, also after reboots. Traffic is not encrypted; use it on networks you trust.
- ZFS replication — scheduled incremental
zfs send | ssh … zfs receiveto another host, old snapshots pruned on both sides, an existing unrelated target is never overwritten. Set up root's SSH key to the target first. - DRS — reads the real host load, Apply live-migrates, fully automated moves machines by itself (one at a time, never the same one twice within 30 min), new machines get a suggested host; settings and rules apply to the whole cluster.
- GFS retention for scheduled snapshots and backups (daily/weekly/monthly/yearly, restic's rules).
- Memory encryption (AMD SEV/SEV-SNP, Intel TDX) is only offered where the host supports it.
Fixes
- HA: a daemon that started while the other members were slow to answer could stop its protected machines at once. Failover itself was tested end to end (host powered off → machine restarted on the survivor, stale copy left off when the host returned).
- Fleet inventory and DRS saw other members as unreachable.
- systemd watchdog and
Type=notifyare active; a busy port now ends virtomird instead of leaving it hanging. - SIEM forwarding settings survive restarts; the dialog shows delivery errors.
- Missing Turkish translations, broken styles in several panels, demo endpoints, AUR package.
Full Changelog: v0.7.2...v0.8.0