Learn DragonFlyBSD - NVMM & Virtualization

#Introduction

In episode 17 you controlled the scheduler and tuned kernel performance. Now we put all that power to one big purpose: running another operating system inside DragonFlyBSD. Episode 18 covers virtualization with NVMM — the built-in type-2 hypervisor that sets DragonFlyBSD apart among modern BSDs.

Remember from episode 0: the safest way to learn is in a VM. Now your position is reversed — you're not the guest, but the host. You'll provide virtual homes for other OSes, give them networking, and create snapshots for their protection.

#NVMM: The Built-in Hypervisor

#What is nvmm(4)

NVMM is a type-2 hypervisor integrated into the DragonFlyBSD kernel. It provides the CPU and memory virtualization primitives used by emulators like QEMU. The advantage: hardware-assisted virtualization with low overhead, without relying on third-party modules like KVM or bhyve.

Checking whether nvmm is available and loaded:

kldstat | grep nvmm
sysctl kern.nvmm

If it's not loaded yet, load the module:

kldload nvmm
echo 'nvmm_load="YES"' >> /boot/loader.conf

nvmm_load in loader.conf loads the module automatically at every boot.

#QEMU Integration

QEMU is the emulator that uses NVMM as its accelerator. Install QEMU from pkg:

pkg install qemu

A QEMU built with NVMM support will use the accelerator when the -accel variable is specified.

#Running VMs

#Running a Linux VM

An example of running a Linux image with NVMM acceleration:

qemu-system-x86_64 -accel nvmm \
  -m 1024 -smp 2 \
  -drive file=/vm/debian.qcow2,format=qcow2 \
  -netdev user,id=net0 -device virtio-net-pci,netdev=net0

-accel nvmm enables acceleration; -m 1024 -smp 2 gives 1 GB of RAM and 2 vCPUs; the disk uses the qcow2 format; networking uses user mode (QEMU's internal NAT).

#Networking: tap and bridge

The user mode is practical for experiments, but for a VM that should appear as a normal host on the network, use tap + bridge:

ifconfig tap0 create
ifconfig bridge0 add em0 add tap0 up
qemu-system-x86_64 -accel nvmm \
  -netdev tap,id=net0,ifname=tap0,script=no,downscript=no \
  -device virtio-net-pci,netdev=net0 \
  -drive file=/vm/debian.qcow2,format=qcow2

A tap interface acts like a virtual cable plugged into the bridge, so the VM gets an address directly from the physical network — the standard pattern for VM servers in a data center.

#VM Management

QEMU VMs are managed through the monitor. Run with the monitor active:

qemu-system-x86_64 -accel nvmm -monitor stdio \
  -drive file=/vm/debian.qcow2,format=qcow2

Inside the monitor, commands like info status, system_reset, and quit give you full control over the VM without closing the host.

#VM Snapshots

Just like HAMMER2, VMs can also be snapshotted. There are two levels:

1. QEMU snapshot — via the qcow2 format:

savevm snap-2026-08-03
loadvm snap-2026-08-03

2. HAMMER2 snapshot — when the VM's disk files live on a HAMMER2 filesystem:

hammer2 snapshot /vm

Combining both gives layered protection: the qcow2 snapshot for VM state (including memory), the HAMMER2 snapshot for the whole directory.

#Closing

In this episode 18 you ran virtualization on DragonFlyBSD: understanding the type-2 hypervisor nvmm(4) and how to load it, integrating QEMU with -accel nvmm, managing VMs with the monitor, setting up networking via tap and bridge, and creating snapshots at both the QEMU and HAMMER2 levels.

Key takeaways:

• NVMM is the built-in type-2 hypervisor; load it with kldload nvmm and persist it in loader.conf.
• QEMU uses NVMM acceleration via -accel nvmm; install it with pkg install qemu.
• Use virtio devices and place VM disks on HAMMER2 for performance and protection.
• Tap + bridge makes a VM appear as a normal host on the network.
• Snapshot at two levels: savevm/loadvm for VM state, hammer2 snapshot for the directory.

In the next episode, episode 19, we combine all your storage knowledge: storage engineering: HAMMER2 production. You'll design a NAS/storage server with multi-volume HAMMER2, manage dedup and encryption, schedule snapshots with cron, monitor with hammer2 status, and use growfs for resizing in 6.2+.