Boot Ubuntu via http/ftp server with pxe(diskless boot)

  • Tutorial

Intro


PXE is a great solution for booting a diskless computer (or a computer without an OS installed). This method is often used for terminal stations and OS mass installation.


Stock ubuntu (16.04) in pxe-mode can mount rootfs only from NFS. But this is not a great idea: any difficulties with the network/NFS server and the user gets problems.


In my opinion, it's best to use other protocols, such as http/ftp. Once booting, you will have an independent system


You should add information about the limits of applicability of the proposed solution and what are the dependencies and restrictions.


Pxe short info


PXE (Preboot Execution Environment) is a special method for booting a computer from the bios / EFI.
How it works (simplified scheme):


  • The computer bios/uefi sends ordinary dhcp request (DHCPDISCOVER)
  • The dhcp server sends a response with the next-server option
  • The computer sends DHCPREQUEST/DHCPINFORM request
  • The dhcp server sends TFTP server address and the filename to upload
  • The computer downloads this file from the tftp server. Its size is limited so, often, it's a bootloader like pxeinux
  • pxelinux reads its own config and downloads Linux kernel using initramfs
  • Linux kernel downloads squashfs with main rootfs
  • switch_root to its squashfs

Keep in mind that TFTP is a slow protocol. It works around UDP with a small block size (512K). Of course, you can increase this value, but this is a way of unstable operation.


A better solution:


  • get bootloader via tftp
  • get kernel (+ initramfs) via tftp
  • get main rootfs squash via http/tftp

How i do it


Steps to solve the task:


  1. Add modules to initramfs
  2. Write my own boot script and add it to initramfs
  3. Make new initramfs
  4. Create squashfs with future rootfs
  5. Setup pxe server
  6. Run it

I used squashfs for rootfs (the simplest way is to create squashfs from installed ubuntu). Overlayfs is necessary to make rootfs writable.


Supported protocols are http/ftp, but you can try to add others via curl/other software.


Customize initramfs


There are 2 places where you can customize initramfs in ubuntu:


  • /etc/initramfs-tools
  • /usr/share/initramfs-tools/

I'll use /usr/share/initramfs-tools. First, I added needed support modules in initramfs:


boozlachu@comp:~$ cat /usr/share/initramfs-tools/modules.d/pxe
overlayfs
squashfs

Next, I wrote a boot script that does all the work:


boozlachu@comp:~$ cat /usr/share/initramfs-tools/scripts/pxe 
#!/bin/bash
mountroot()
{
        maxTryCount=5
        squashfsFile="/tmp/rootfs.squashfs"
        squashfsMountPoint="/mnt/ro"
        tmpfsMountPoint="/mnt/rw"
        overlayfsUppderDir="$tmpfsMountPoint/upper"
        overlayfsWorkDir="$tmpfsMountPoint/work"
        overlayfsDir="/mnt/overlayfs"
        tryCount="1"

        # run udevadm
        wait_for_udev 10

        # parce kernel cmdline args. rooturl needed
        for x in $(cat /proc/cmdline); do
                case $x in
                rooturl=*)
                        export rooturl=${x#rooturl=}
                        ;;
                maxTryCount=*)
                        export maxTryCount=${x#maxTryCount=}
                        ;;
                esac
        done

        log_begin_msg "Loading modules" 
        modprobe squashfs || panic "can't modprobe squashfs"
        modprobe af_packet || panic "can't modprobe af_packet"
        modprobe overlay || panic "can't modprobe overlayfs"
        log_success_msg "modules loaded"

        log_begin_msg "Configure network"
        configure_networking || panic "Can't configure network"
        log_success_msg "Network configured"

        log_begin_msg "Download rootfs"
        while [ ! -f ${squashfsFile} ] && [ ${tryCount} -le ${maxTryCount} ]; do
                wget ${rooturl} -O ${squashfsFile} || log_failure_msg "Can't download rootfs, count ${tryCount}"
                tryCount=$(( ${tryCount} + 1 ))
                sleep 0.5
        done

        if [ -f ${squashfsFile} ]
        then
                log_success_msg "Rootfs downloaded"
        else
                panic "Can't download rootfs"
        fi

        log_begin_msg "Mount rootfs"
        mkdir -p $squashfsMountPoint
        mount -t squashfs -o loop $squashfsFile $squashfsMountPoint || panic "Can't mount rootfs"
        log_success_msg "Rootfs mounted"

        log_begin_msg "Mount tmpfs"
        mkdir -p $tmpfsMountPoint
        mount -t tmpfs none $tmpfsMountPoint || panic "Tmpfs mount failed "
        log_success_msg "Tmpfs mounted"

        log_begin_msg "Mount overlayfs"
        mkdir -p $overlayfsUppderDir $overlayfsWorkDir $overlayfsDir
        mount -t overlay overlay -o lowerdir=$squashfsMountPoint,upperdir=$overlayfsUppderDir,workdir=$overlayfsWorkDir $overlayfsDir \
                || panic "Overlayfs mount failed"
        log_success_msg "Overlayfs mounted"

        log_begin_msg "Move tmpfs and squashfs to new root"
        mkdir -p $overlayfsDir/$tmpfsMountPoint $overlayfsDir/$squashfsMountPoint
        mount --move $squashfsMountPoint $overlayfsDir/$squashfsMountPoint  || panic "squashfs move failed"
        mount --move $tmpfsMountPoint $overlayfsDir/$tmpfsMountPoint  || panic "tmpfs move failed"
        log_success_msg "Tmpfs and squashfs moved"

        log_begin_msg "Move overlayfs to new root"
        mount --move $overlayfsDir ${rootmnt}  || panic ""
}

The script has a lot of messages for understanding how it works.


After the modules and script, add your need to generate new initramfs:


boozlachu@comp:~$ sudo update-initramfs -c -k all

Creating squashfs


The simplest method:


  • install ubuntu on drive
  • boot from LiveCD
  • create squashfs from the installed system

I don't recommend this way for production since you'll have a very large squashfs (not the best idea for pxe)!


Setup bootloader, squashfs, and pxe server


I use pxelinux as a pxe bootloader. It's an easy way. My pxe servers are Debian 10, tftp-hpa,dhcpd, and lighttpd.
I’ll omit the installation details, but I’ll show the important info.


TFRP server file struct (/srv/tftp is root dir fot tftp-hpa):


root@debian:/srv/tftp/ubuntu# tree /srv/tftp/
/srv/tftp/
└── ubuntu
    ├── firmware.sq
    ├── initrd
    ├── ldlinux.c32
    ├── libcom32.c32
    ├── libutil.c32
    ├── menu.c32
    ├── pxelinux.bin
    ├── pxelinux.cfg
    │   └── default
    ├── vesamenu.c32
    └── vmlinuz

  • firmware.sq is squashfs with rootfs
  • *c32 are files for pxelinux
  • vmlinuz is kernel
  • initrd is initramfs(which i rebuild earler)
  • pxelinux.bin — main pxelinux file
  • default is config for pxelinux

Bootloader config:


root@debian:/srv/tftp/ubuntu# cat /srv/tftp/ubuntu/pxelinux.cfg/default 
ui menu.c32
timeout 30
default ubuntu_pxe 
font UniCyr_8x16.psf
menu title PXE Special Boot Menu
menu color tabmsg 37;40      #80ffffff #00000000
menu color hotsel 30;47      #40000000 #20ffffff
menu color sel 30;47      #40000000 #20ffffff
menu color scrollbar 30;47      #40000000 #20ffffff

LABEL ubuntu_pxe
menu label  Run ubuntu pxe 
        kernel vmlinuz 
        append initrd=initrd rooturl=http://192.168.56.2/ubuntu/firmware.sq boot=pxe maxTryCount=10

It’s impotant to set the correct kernel parameters:



And the last one is the dhcp config:


root@debian:/srv/tftp/ubuntu# cat /etc/dhcp/dhcpd.conf
subnet 192.168.56.0 netmask 255.255.255.0 {
        range 192.168.56.10 192.168.56.45;
        option routers 192.168.56.2;
        option domain-name-servers 192.168.2.1 ;
        option broadcast-address 192.168.56.255;
        default-lease-time 3600;
        max-lease-time 7200;
        # Important! Set bootloader file
        filename "ubuntu/pxelinux.bin";
    }

The extended variant (if the dhcp and tftp servers placed on different machines) requires the next-server option for dhcp.


Conclusion


This article shows you how to change the boot mode of ubuntu without any difficulties. Use it as information and write your solutions. This can be a system in the form of firmware (with squashfs), pxe, or another solution.

Similar posts

Ads
AdBlock has stolen the banner, but banners are not teeth — they will be back

More

Comments 3

    +1

    Such an awesome article, thank you!
    Have you tried LTSP? — it is doing the same but having ready scripts for many routine tasks.

      0
      I guess LTSP is a complete solution for network boot. But in my practice wa are don works without it.
      My article is more than pxe tutorial. I tried to show how to customize ubuntu boot to use ubuntu as firmware from squashfs. PXE is only one of variant
        0
        In addition, LTSP uses the nfs / nbd protocol. I don't like when network connection dependencies work on my system.
        HTTP and FTP allows booting ALL systems from rootfs to my client. Then my system will be offline.

      Only users with full accounts can post comments. Log in, please.