I'm seeing a nasty virtio network regression with the 6.14 Linux kernel on Fedora 41, 42, and OpenSUSE Tumbleweed. "bad gso: type 4" and http traffic crawling to ISDN speeds. Specifically http/s, somehow iperf tcp performance is fine, but dnf, curl, wget, etc, are affected. Trying to gather what info I can and I'll file a bug about it soon unless someone beats me to it.

In the wee small hours i decided to give #GnomeBoxes another chance as a possible alternative. Chose to install via #flatpak, coz my Host is #ArchLinux #KDEPlasma & i did not wish to risk possible gnome vs plasma file conflicts.

It did a fine job of installing an #ArchLinux VM [which i completed using the latest #ArchInstall v3.0.3], so then in an act of gross stupidity i ridiculously chose to see if it could handle a win10 VM via a 3-yr old ISO i have lying around. As i abandoned windoze "for real" in 2014, i had forgotten the many hours of time-wasting bullshit involved, at the end of which instead of getting something decent, you just get... windoze, ugh.

Anyway, so far it seems that GnomeBoxes is going ok, which makes a nice change as all prior tests of older versions failed to work properly. Doubt though atm that it'll persuade me to give up on #qemu / #kvm / #libvirt / #VirtManager. GB has no facility to Suspend VMs, so they need to be fully shutdown, which is a drag. Also, the FP version does not support USB transfer, which is a bugger, & so far i've not made #SharedFolders work [but think i know why].

Ansible k3s on vms 1.2.0 and delegated_vm_install 2.0.3 released

https://stafwag.github.io/blog/blog/2025/04/06/ansible-k3s-on-vms_and_delegated_vm_install_released/

I prepared a few update releases of some ansible roles related to provisoin virtual machines with libvirt over the last weeks.

Mainly clean up releases and makes sure that everything works on different GNU/Linux distributions out of the box.

#ansible #cloudinit #kvm #libvirt #linux

#stafwag @stafwag

How Do I Properly Install KVM on Linux

https://sysguides.com/install-kvm-on-linux

The Kernel-based Virtual Machine (KVM) is a Linux hypervisor that supports full virtualization. When you install KVM on Linux, your Linux distribution is transformed into a Type-1 hypervisor, allowing you to run virtual machines at near-host machine speeds.

A Type-1 hypervisor, also known as a bare-metal hypervisor, interacts with the underlying machine hardware directly rather than through an operating system.

In this article, I'll show you how to properly install KVM on Linux distributions such as Fedora, Rocky, Ubuntu, Debian, and Arch Linux.

Table of Contents

1. Overview of Key KVM Components 1.1. KVM Kernel Modules 1.2. QEMU 1.3. Libvirt
2. Check Virtualization Support
3. Install KVM on Linux Distributions
4. Install VirtIO Drivers for Windows Guests
5. Enable the Modular libvirt Daemon
6. Validate Host Virtualization Setup
7. Optimize the Host with TuneD
8. Configure a Network Bridge
9. Give the User System-Wide Permission
10. Set ACL on the Images Directory
11. Conclusion
12. Watch on YouTube

In one of your recent stream VODs, @tomlawrence, someone asked, whether they could run #ZFS on #LUKS - i can answer that; YES*, with an *asterisk.

I did this for quite some time, until i've decided that it's rather inconvenient to type in my password on every reboot. Now, i'm running LUKS on ZVOLs, in #Ubuntu / #qemu / #libvirt.

It's a small home server, and i need a few "privacy insensitive" VMs to auto-start after power-fail.

All one needs is a block dev, zpool create, done! …technically

Tumbleweed keeps rolling strong in 2025! January brought major updates like #GIMP 3.0 RC & #libvirt 11.0.0. #KDE Plasma 6.2.5 & more also arrived in #snapshots. #openSUSE https://news.opensuse.org/2025/01/29/tw-monthly-update-january/

Just made a major shake-up of my infrastructure… having cleared my 16-core workhorse of VMs, I've replaced the Debian Buster OS on there, with a shiny new #AlpineLinux OS.

My Mastodon instance is currently running on the 8-core node I resurrected last week and finalised yesterday.

I'm finding a few things are different with #libvirt under Alpine, notably support for SPICE seems lacking out-of-the-box (so VNC console only) and I've still got oddities with RBD pools, but VMs still successfully connect to RBD back-ends anyway.

It's nice to have nodes running up-to-date OSes. I might need to fine-tune some things, but generally I'm happy with this.

Well… fun and games today trying to get the newly refurbished node up and running proper.

I figured I'd start by getting LACP configured. Simple enough.

Enter one managed switch with a not-that-well thought out UI. It took far too long for me to figure out why VLANs weren't working… then in the midst of this, I managed to really upset the apple cart regarding the network switching.

I've only just managed to get that node onto the DIN rail where it belongs and finalised its network config.

I'm now having fun and games with getting #libvirt to talk to #ceph. It has two pools defined, and it starts one just fine… but the other (as it happens, the one that has 90% of my data in it), refuses to start, with an "unknown" error.

https://www.spinics.net/linux/fedora/libvirt-users/msg14421.html gives the full detail of my prior debugging efforts… but basically this node is doing the exact same thing.

Quite frustrating. I'll have to leave the repairs of the final node until later, maybe next week-end.

#FediBrain moving to libvirt but still using Vagrant - only vagrant 2.4.0-1 seems to work for me - Older versions cannot download boxes, newer versions doesn't recognize libvirt. What the f... did I do wrong or is it just a mess?
#libvirt #vagrant #version #WhatAMess

In case this piece of knowledge is useful to someone: if you install a guest OS in #GNOMEBoxes while the host computer's network is disabled (to dodge Windows' stupid SaaS-style "online accounts as system accounts" enshittification push, for example), there is a bug where GNOME Boxes will not be able to automatically set up the virtual bridge for the guest OS to be reachable from your LAN/host via the network: https://gitlab.gnome.org/GNOME/gnome-boxes/-/issues/1122

Screw Broadcom, but I've just wasted an hour trying to make #libvirt/virtsh/virt-manager work. I need a working hypervisor so I'm going back to VMWare Workstation on Linuux.

(system vs user sessions, bridged networking not working for me, silly error messages re swtpm and more)

I've been using #virtiofs to access the host filesystem in my #guix #kvm #libvirt #VMs - it's very nice. I have it mount the directory automatically in my guix-system definition:

(file-system
(mount-point "/home/la/workspace")
(device "workspace")
(type "virtiofs")

Guix also supports the 9p filesystem and others.

well, I figured out why you shouldn't run immutable fedora: you can't use libvirt. Well, you can *sort of*, if you don't care about virtual networking or device passthrough or ... any of the other reasons why you'd use libvirt vs. just running in a container.

Anyway, this seems like a pretty goddamn huge problem but ... apparently nobody cares? Or at least, there appears to be no significant effort to actually make it work. So I guess I get to wipe my laptop and start over again.

Fuck

How do I make #libvirt request & use prefix delegation to give my VMs globally routable, externally visible #IPv6 addresses? (I don't want bridging, I specifically want routing/forwarding with prefix delegation, just like any router does.)

It seems to default to giving VMs no IPv6 connectivity at all, not even outbound (how is this acceptable?), and seemingly the recommended (the only supported?) way to get IPv6 is to use NAT?

I almost never fully exercise this 10+ year old xeon, but every once in a while I'll do some multitasking and I thought I'd share a screenshot to commemorate. I could hardly tell it was chugging away at #liveslak while I spun up a win10 vm to run some weekly reports. #slackware #libvirt

Using vagrant with #libvirt on #fedora 40
https://ludovic.hirlimann.net/2024/04/using-vagrant-with-libvirt-on-fedora-40.html