Easy PXE boot testing with only HTTP using iPXE and libvirt

Update: A future post explains how to do this even easier without PXELINUX.


Occasionally I have a need to test out a PXE install workflow. All of this is super easy if you have a permanent PXE infrastructure you maintain which traditionally has consisted of DHCP, TFTP and HTTP/FTP servers. What if I just have my laptop and want to test something in a VM? It turns out it’s pretty easy to do using libvirt and a simple http server.

In the steps below I walk through setting up libvirt to point to a web server for PXE booting that has been set up with all the files needed for testing out a PXE install workflow.


Libvirt uses iPXE as the firmware for network booting in the network interfaces for VMs. This firmware has the ability to retrieve the pxelinux boot file, kernel, and initrd over HTTP, which means that we aren’t required to set up a TFTP server any longer.

Web Server Setup

NOTE: I’m executing all of these steps on a Fedora 29 host.

The first thing we can do is make a new directory and switch to it.

$ mkdir pxeserver && cd pxeserver

Next we’ll grab pxelinux.0 and ldlinux.c32 files from the syslinux packages within Fedora:

NOTE: You can execute this step in a container if you do not want to install packages on the host.

$ dnf install -y syslinux-nonlinux
$ cp /usr/share/syslinux/{pxelinux.0,ldlinux.c32} ./

Next I’ll use the install media for Fedora 29 server to create an install tree which can be served by our web server. In the commands below I’m just loopmounting the ISO image on top of the Fedora-Server-dvd-x86_64-29-1.2.iso directory. I could have also just used rsync to copy the files out of the ISO too. Either way works.

$ mkdir Fedora-Server-dvd-x86_64-29-1.2.iso
$ sudo mount -o loop /var/b/images/Fedora-Server-dvd-x86_64-29-1.2.iso ./Fedora-Server-dvd-x86_64-29-1.2.iso/
mount: /var/b/shared/pxeserver/Fedora-Server-dvd-x86_64-29-1.2.iso: WARNING: device write-protected, mounted read-only.
$ ls Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/
initrd.img  TRANS.TBL  vmlinuz

You can see from that last ls that the kernel and initrd for pxe booting are under Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/. We’ll now create a default pxelinux configuration file and specify the location of the kernel and initrd.

$ mkdir pxelinux.cfg
$ cat <<EOF > pxelinux.cfg/default
DEFAULT pxeboot
LABEL pxeboot
    KERNEL Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/vmlinuz
    APPEND initrd=Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/initrd.img console=ttyS0 inst.ks=

In the configuration file we do specify console=ttyS0 so we can view output from the install on the serial console. We also specify the location of a kickstart file which I’ll create now:

$ cat <<EOF > kickstart.cfg
url --url
rootpw --plaintext foobar
services --enabled="sshd,chronyd"
clearpart --all
autopart --type lvm

This is just an example kickstart file and is as simple as possible.

Now our directory structure looks like this:

$ tree -L 2
├── Fedora-Server-dvd-x86_64-29-1.2.iso
│   ├── EFI
│   ├── Fedora-Legal-README.txt
│   ├── images
│   ├── isolinux
│   ├── LICENSE
│   ├── media.repo
│   ├── Packages
│   ├── repodata
│   └── TRANS.TBL
├── kickstart.cfg
├── ldlinux.c32
├── pxelinux.0
└── pxelinux.cfg
    └── default

We’ll go ahead and start the web server in this terminal:

NOTE: You may need to poke a hole in your firewall for port 8000. You can use a command like sudo firewall-cmd --add-port 8000/tcp.

$ python3 -m http.server 
Serving HTTP on port 8000 ( ...

Libvirt network DHCP setup

In another terminal we’ll edit the network configuration for the default libvirt network to tell it to serve out a boot dhcp option that tells clients where a network boot file is that can be used to network boot the mahcine. We do this by adding a <bootp file=''/> line to the <dhcp> XML element.

NOTE: Please bring down all VMs on your default libvirt network before executing this next step.

The steps for bringing down, editing, and re-starting the libvirt network look like:

$ virsh net-destroy default
$ virsh net-edit default
$ virsh net-start default

When all is done my final XML looked like:

  <forward mode='nat'/>
  <bridge name='virbr0' stp='on' delay='0'/>
  <mac address='52:54:00:aa:da:73'/>
  <ip address='' netmask=''>
      <range start='' end=''/>
      <bootp file=''/>

PXE Booting!

Finally I should be able to start a virtual machine and have it grab the PXE configuration I have defined and is being served from the python web server. The command I use to start a VM for this looks like:

$ virt-install --pxe --network network=default --name pxe --memory 2048 --disk size=10 --nographics --boot menu=on,useserial=on

In the serial console I can see fact that the pxelinux.0 file and the vmlinuz and initrd.img were retrieved:

SeaBIOS (version ?-20180724_192412-buildhw-07.phx2.fedoraproject.org-1.fc29)
Machine UUID af153013-d6ef-421b-b0f1-1bbfd88cd7e9

iPXE (http://ipxe.org) 00:02.0 C100 PCI2.10 PnP PMM+7FF91200+7FED1200 C100

Booting from ROM...
iPXE (PCI 00:02.0) starting execution...ok
iPXE initialising devices...ok

iPXE 1.0.0+ -- Open Source Network Boot Firmware -- http://ipxe.org

net0: 52:54:00:59:73:4c using rtl8139 on 0000:00:02.0 (open)
  [Link:up, TX:0 TXE:0 RX:0 RXE:0]
Configuring (net0 52:54:00:59:73:4c).................. ok
net0: gw
net0: fe80::5054:ff:fe59:734c/64
Next server:
Filename: ok
pxelinux.0 : 42780 bytes [PXE-NBP]

PXELINUX 6.04 PXE  Copyright (C) 1994-2015 H. Peter Anvin et al
Loading Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/vmlinuz... ok
Loading Fedora-Server-dvd-x86_64-29-1.2.iso/images/pxeboot/initrd.img...ok
Probing EDD (edd=off to disable)... ok

The machine continues on booting and completes the kickstart!


So we were able to PXE boot without TFTP and by re-using the DHCP server that is already built into libvirt. Even though we took shortcuts this should work with any network card that supports iPXE if you can configure the DHCP server in the network to serve out the filename to the nodes.

Happy New Year All!