If you work a lot with your Raspberry Pi, you probably have burned one or another SD card. While SD flash cards are great for storing photos or music, they suck at being written to very often. This is exactly the case if you run a whole operating system on them. Hundreds of thousands of read and write operations wear your SD cards out and if you do not have a solid backup concept you have probably lost some data, too.

In my home, a couple of systems are running on the cheap, embedded hardware by the Raspberry Pi Foundation because they are both small and power-saving at a very affordable price point. And so are the accessories like cases and add-on-boards. At the time of writing, these are the jobs that are done by my Raspberries:

  • SSH server (as gateway and tunnel from the internet into the LAN)
  • Monitoring
  • OpenHAB smart home controller
  • TV Server
  • Backup
  • Ambilight


When living alone, having all this run on SD card-based Raspberries is fine and you know what to do when something breaks. As soon as you have a partner to share your home with, you probably need to agree on an SLA to ensure a high WAF:

SLA (Service Level Agreement):

If you want to have a smart home controlled by a computer, you have a very short acceptable “TTR” if something breaks. Our Time-To-Recover is usually as short as 1 day. Honestly, it is more like same-day. Do not go to sleep until it is fixed.

WAF (Wife Acceptance Factor):

If you stick to the SLA, you are free to do all this crazy stuff; your partner accepts it and maybe even likes it.


And that is basically the reason I am writing this. It is completely OK if the ambilight stops working, it is also acceptable if the backups do not run for some days. But if the smart home controller fails, … well, you better have a backup plan. You do not want to deal with someone who spent two hours trying to turn the lights up or the television on and ultimately failed because it simply does not work anymore without the central controlling instance.

So we should reduce these outages to a minimum:

The Objective

  • Increase device uptime / reduce device outages
  • save data on a central storage for easy backups or snapshots


There are so many other things that can fail. The processor may overheat and hang, forcing a reboot by the watchdog, the power consumption could be too high and disable the USB bus (including the Ethernet Port), the cheap phone charger you use for powering may fail as well, but the single biggest problem is – without a doubt – the SD card.
So today I am showing you how simple it is to boot and run your Pi completely from the network.


The Requirements

  • always-on network storage with NFS server software
  • stable network connection to the storage (wired whenever possible)
  • 16+ MB SD card for the Raspberry Pi bootloader and config file that points to your network storage
    (Note: Starting with the Raspberry Pi 3 there are ways to modify the internal bootloader to boot from network without an SD card. I am not going to cover this here; this method will work with all the Raspberry Pi versions)
  • 3 beers of time

Everything I do is based on a Linux computer (I use Arch Linux and Manjaro) but you can also do this on OS X and maybe on Windows with a lot of tools. I really recommend you to use a Linux computer or a linux VM.

The Network Storage

You need a network storage capable of running an nfs-server. You can even use a Raspberry Pi as sever but that would not really solve the problem here, would it? 😉

Ensure the NFS service is globally enabled

Most commercial NAS systems support NFS and so does my Seagate BlackArmor NAS 4000. I highly recommend to create a seperate share for the filesystems of your Raspberries to be able to add stricter permissions later.

You may need to enable the NFS service first, because a lot of people do not need it and use CIFS only.

Create a share where you will put the filesystems for your Raspis

The resulting config, that my NAS generates, looks like this:

~ $ grep rootfs /etc/exports
/nfs/rootfs *(rw,async,insecure,no_root_squash,no_subtree_check)

Please note that the options rw, no_root_squash and no_subtree_check are obligatory for what we want to achieve. For a detailed description of all possible options, please consult the man page on your system or here.


The Extraction Of The SD Card Image Content

You have your NFS Server up and running and configured.

I already stumbled across some distributions of Raspberry Pi SD card images, that depend on and wait for partitions on the local sd card – even if you boot them from NFS. If you see your Raspberry Pi failing at boot, because it waited for a local partition or device for too long, simply flash the sd card image onto that sd card like you would normally. Then, change the settings to have it boot over NFS. Your local SD card still will not be touched, but the boot scripts and dependencies will work now.

This topic can be split into two subtopics: Using a freshly downloaded image to start from scratch, or converting a normal SD-card-based Raspberry Pi to a NFS-based one. Skip to Convert SD-Card-Based Raspberry Pi To NFS if you already have an SD card image, which you want to convert.


Create A RootFS From Scratch (New Download)

Now you can download any SD card image like you would do normally. But instead of flashing it onto an SD card, you will mount it and copy the content to your NFS share.

In this example I will download the official Raspberry Pi Foundation Raspbian Stretch Lite Image over the Torrent protocol on my Manjaro Linux machine.

[[email protected] Downloads]$ transmission-cli https://downloads.raspberrypi.org/raspbian_lite_latest.torrent
2017-09-07-raspbian-stretch-lite.zip: State changed from "Incomplete" to "Complete"

Be nice and give something back to the torrent community by leaving this open seeding (uploading) for the other downloaders. Meanwhile in another terminal window:

Check the checksum to check if the data was downloaded successfully and was not modified:

[[email protected] Downloads]$ sha256sum 2017-09-07-raspbian-stretch-lite.zip 
bd2c04b94154c9804cc1f3069d15e984c927b750056dd86b9d86a0ad4be97f12 2017-09-07-raspbian-stretch-lite.zip

The checksum equals the checksum from the website. Great! Let’s unpack it and have a look inside the partition table. It is a complete device image, so it contains everything from (or for) the SD card, including partition table.

[[email protected] Downloads]$ unzip 2017-09-07-raspbian-stretch-lite.zip 
Archive: 2017-09-07-raspbian-stretch-lite.zip
 inflating: 2017-09-07-raspbian-stretch-lite.img
[[email protected] Downloads]$ fdisk -l 2017-09-07-raspbian-stretch-lite.img 
Disk 2017-09-07-raspbian-stretch-lite.img: 1.7 GiB, 1854590976 bytes, 3622248 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
                                ^^^-----<<< write this down! <<<-------
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0x11eccc69

Device                           Boot Start  End      Sectors Size Id Type
2017-09-07-raspbian-stretch-lite.img1 8192   93813    85622   41.8M c W95 FAT32 (LBA)
2017-09-07-raspbian-stretch-lite.img2 94208  3622247  3528040 1.7G 83 Linux
                                      ^^^^^-----<<< and this! <<<-------

In the output of fdisk -l you see some important information. As we are not interested in putting the boot partition onto our NFS server (but the rootfs partition, the second partition) instead, we need to know the offset where the interesting partition starts. To get this offset you simply multiply the sector size (usually always 512 bytes) with the start sector of the second partition (both are marked above).

[[email protected] Downloads]$ echo $(( 512 * 94208 ))

Let’s create a loopback device with this information so we can mount it and steal the data from it.

[[email protected] Downloads]$ sudo losetup -f --show -o $((512*94208)) 2017-09-07-raspbian-stretch-lite.img 

You need to be root to create loopback devices. -f found us a free loop device, -o took the offset we calculated and –show made sure we get to know which loop device losetup used. Now we can go and mount it like any other partition.

[[email protected] Downloads]$ mkdir /tmp/fresh-pi-rootfs
[s[email protected] Downloads]$ sudo mount -v /dev/loop1 /tmp/fresh-pi-rootfs
mount: /dev/loop1 mounted on /tmp/fresh-pi-rootfs.
[[email protected] Downloads]$ cd /tmp/fresh-pi-rootfs
[[email protected] fresh-pi-rootfs]$ ls
bin boot dev etc home lib lost+found media mnt opt proc root run sbin srv sys tmp usr var

Now let’s copy that data to your NFS share. Refer to the code snippet in Convert SD-Card-Based Raspberry Pi To NFS, but copy from /tmp/fresh-pi-rootfs instead of /.

Afterwards, burn the SD card image to your SD card like you would normally. This ensures that you have the proper boot partition (the first partition in the partition table) on your card. If you feel fancy or have a tiny SD card, you can also extract the partition data like we did above and put it onto a fresh FAT partition on the SD card with the same start sector.


Convert SD-Card-Based Raspberry Pi To NFS

If you already have a running Raspberry Pi, you probably do not want to rebuild everything from scratch but move your existing data to the network share. All you need to do is to log into your Pi, install rsync (or any other application that reliably copies filesystem attributes like scp), and copy all your stuff over. Do not forget to stop your applications to have them in a defined (stopped) state.

[email protected] ~ $ mkdir /tmp/rootfs
[email protected] ~ $ mount nas:/nfs/rootfs /tmp/rootfs
[email protected] ~ $ rsync -Phax --numeric-ids / /tmp/rootfs/openhab

There are about 31.000 files to copy, so this might take a while. Go brew some coffee or grab a beer.


The Configuration

All your data are belong to the NFS share. The last step is to tell the Raspberry Pi to boot from the share instead of the local SD card. One widespread method that works on almost every linux device is to modify the kernel command line. This is usually done in the bootloader (Syslinux,  Grub, U-Boot). The Raspberry Pi got a configurable U-Boot bootloader starting with version 3. In the versions before you have to have a partition on your SD-Card, containing the boot files and a file called “cmdline.txt” that contains the kernel command line. This also works with Version 3 and newer, if you do not want to bother with U-Boot and you are fine with using an SD card for that.

Modify the “cmdline.txt” file on the boot partition. You can do this directly on your Raspberry Pi, or by plugging the SD card into another computer. When using another computer, you’ll find the file on the first Partition of the card, usually called “boot” partition. When editing the file directly on the Pi, it is located at “/boot/cmdline.txt”.

This is an example cmdline.txt I use in production:

dwc_otg.lpm_enable=0 console=serial0,115200 console=tty1 root=/dev/nfs nfsroot= ip=dhcp elevator=deadline

This is basically the default cmdline.txt with additions of the root, nfsroot and ip options. If you want to define a static IP address, use something like



Please keep in mind, that there must only be exactly one line in the cmdline.txt file. No comments, nothing else. Just one line.


The First Boot

Save it, unmount the boot partition (Apple and Microsoft call this “eject securely” or similar), put the SD card into your Pi and fire it up. If you have a screen attached, you will see the communication with the DHCP server and the handover to the NFS filesystem. This usually takes only a second. My boot time increases by 50% which is most likely the case because of my very old and slow NAS. Results may vary.

Congratulations, you just separated hardware from software resources and made your IT a lot more agile. Now, if your Pi explodes, you can just swap it with a new one from you Raspberry-Pi-drawer and be up again in no time. I am not responsible for exploding Pis, though.


The Remaining Questions

People asked me about this a lot so maybe this is interesting for you as well:

Q: Wait! What happens, if my NAS crashes, or is offline in some other way?
A: As soon as the NFS server doesn’t answer anymore, your devices will just stop completely. The root filesystem is essential so your Raspberry Pi will just wait until it is reachable again. It will not crash and it won’t reboot or panic. It just waits until your server is back. I think this is quite nice.


If you have any other questions, feel free to ask in the comments section below!


Read Also

Leroy Förster

Leroy Förster

Leroy is a heavy Linux user and smart home enthusiast born in 1991 and living in Hamburg, Germany. Other hobbies include tinkering with open source hardware and composing mostly electronic music.



Matt · 25. January 2019 at 17:01

Thank you so much for this, took me 2 days to implement on my setup but was a massive help.

issues i had were my nas did not allow root access over the network so when rsync was run it could update file permissions as it created them. After that the pi would boot into emergency mode, i noticed it was booting as readonly. added rw to the end of the cmdline.txt and boom….done.

only think i dont like is if i change my item files directly on the NAS openhab doesnt detect the file change and doesnt refresh. not a major issue but would like to resolve it.

dont suppose you had the same issue?

Thanks again,

Leroy Förster

Leroy Förster · 25. January 2019 at 22:23

Hey Matt, very glad to hear my post was useful to you.

If using NFS, the server option “no_root_squash” allows your NFS client, in that case, your openHAB RaspberryPi, to create files as the root user. When “no_root_squash” is NOT set, or when setting “root_squash”, all files created as root get squashed to a special user to prevent leaking root data of other users in multi-user NFS environments. Is “no_root_squash” properly set on your NFS server AND client?

Regarding your issues with openHAB not noticing the file changes: openHAB registers inotify on the kernel for its config folders, so that the kernel notifies openHAB if any of the files changed. This should work well over NFS, too, but it works the other way round. So you will need an NFS or CIFS server on your Pi because the inotify events get emitted by the server and thus on the server system. This is your NAS and nothing is monitoring the files on your NAS.

It tested this on my setup by creating a folder as root directly on the nas NFS server:

~ $ cd /DataVolume/rootfs/openhab/tmp/
/DataVolume/rootfs/openhab/tmp $ mkdir folder_created_on_nas
/DataVolume/rootfs/openhab/tmp $ cd folder_created_on_nas/

Then on the Raspberry Pi I installed the `inotify-tools` Debian package and used the helper tool inotify-wait to monitor the folder. I then created files and folders from the NAS:

[openHAB Pi]
[email protected]:/tmp/folder_created_on_nas# inotifywait -m -r .
Setting up watches. Beware: since -r was given, this may take a while!
Watches established.

None of the events for the created files or folders were fired by the kernel on the NFS client (openHABian Pi). This is the reason that openHAB does not recognize the change.

Because I use openHABian and it ships with a pre-configured CIFS server which exports /etc/openhab2, I took the easy way and connect directly to the CIFS server on the Pi, instead of connecting to the NAS and changing the files there.

When using an NFS or CIFS server on the Pi itself and using that to change your config, you can use all the features that inotify and openHAB provides.

TL;DR: Install NFS or CIFS SERVER on the device running openHAB and use that to change your openHAB configuration.

If you get stuck, you are always welcome to get more help here.


Wayne O'Neil · 19. September 2019 at 06:16

What’s the speed difference running on NFS vs. SD?

Also I was thinking of making an RPi NFS server.. You had mentioned that this defeats the purpose, but since the NAS disk is local to the RPi, I can still boot off a partition on it, correct?

    Leroy Förster

    Leroy Förster · 19. September 2019 at 09:04

    I don’t have any exact measurements, but using my relatively slow NAS (might be comparable to a Raspberry Pi 3 NAS) booting the other Raspberry Pis from it is significantly slower. Booting takes about twice the time compared to a normal SD card.

    You can export a partition that you boot from, yes. But you obviously can not boot the NFS server from itself, before it is running. If you use the RPi as an NFS Server, it must boot from a local disk or SD card.


Jay · 26. September 2019 at 17:22

I would like to implement this solution on my pi2. I have some new pi4s waiting to be setup but before that I wanted to tackle the network boot option.

I have a router that is using ddwrt and has a usb port for NAS. The ddwrt firmware itself has a baked-in option for proftpd which then uses the usb attached storage.

Is it possible to have the pi boot from the proftpd service or is this procedure strictly NFS?

    Leroy Förster

    Leroy Förster · 29. September 2019 at 19:32

    It’s only possible using NFS, not with FTP. FTP was made for a completely different purpose and thus lacks important file system features such as scheduling or locking.

    A few years ago I was using DD-WRT as well but then switched to OpenWRT for greated flexibility. It also has NFS Server packages ready to use.


thefa · 27. October 2019 at 19:10


Thanks a lot for the detailed info you provide here. I’m wondering if you could comment on the UID/GID mapping topics… Did you have to create users and groups with matching UID and GID to make this solution work? Or is it ok to have the NAS operate with numeric values for UIDs and GIDs and everything is ok this way?

    Leroy Förster

    Leroy Förster · 4. November 2019 at 11:03

    Using the numeric IDs works just fine. You don’t have to create users and groups with these IDs on the NFS server. The server doesn’t bother if there is a user or not, only the client does.

    This also means that if you require privacy of your files, you must still ensure that client “Alice” has a different UID/GID than “Bob”, but this is a classic NFS topic and does not only apply for the raspberry pi.

    So – everything should be fine and you should be good to go 🙂


Benoni · 30. November 2019 at 18:21

Perfect write up, thank you.


Tony · 5. July 2020 at 13:46

Thanks so much for the article, it’s helping me moving my pi it’s rootfs to NFS 🙂

I copied over the existing rootfs using the rsync command (rsync -Phax –numeric-ids / /mnt/nfs/rpi-root) but my pi is unable to find the rootfs on nfs; error: “Failed to start Remount Root and Kernel File Systems”.

Took a picture of the errors during the boot process: https://imgur.com/gallery/mhUp2RP

Any ideas..?

Leave a Reply

Your email address will not be published.

1 × four =

This website is using Google Analytics. Please click here if you want to opt-out. Click here to opt-out.