For a while I have been using Samba to remotely connect Windows computers to my Linux computers and one Linux file server. I can even connect my Linux laptop to my Linux server via Samba. But, recently I bought a Raspberry Pi and I got interested in using NFS for three reasons.
- The Raspberry Pi will be running 24/7 and I would like the option to automount the home folder and others from it on my Linux computers when I start them up.
- I would like to mount some folders on a server (“big” file storage Linux server) that I can start remotely to the Raspberry Pi so that it will act like expanded storage on the Raspberry Pi. Then I can start the “big” server remotely and mount the folders on the Pi and use the Pi as a proxy. So it is connected to the web and I can navigate to folders that are on the big server via a connection to the Pi with NFS mounted directories.
- It would make it easy to backup Linux machines, including the Pi to the big server periodically. Years ago I thought about NFS for mounting folders for backup but I was pretty happy using a scripted FTP system for backups, so I shelved implementing NFS mounts back then.
Implementing NFS was a lot easier than I thought it would be. It was actually much easier than getting Samba to work the way I wanted it to.
The first step ( in my opinion) is to have the machines that you will mount directories from and to on static IP address. On a home network it really does make it easier to have all the machines other than guest machines on static IP’s. This can be done either by setting the machine to have a static IP. Or it may be possible depending on the router, for the router to be configured to hand out the same IP address to a machine with a specific MAC ID. Effectively the results are the same.
Static IP’s are useful as the actual IP addresses will be listed in the export file. It may be possible to use names, however this depends on how DNS is handled on your network. Using the actual IP addresses will make initial setup nearly foolproof. Also an easy way to use names on any machine is to add the static IP’s and names of the machines on the network to the /etc/hosts file.
Install NFS Support
To install support for NFS on the machines run….
sudo apt-get install nfs-kernel-server
Exports File on Server
Modify the servers /etc/exports file to suit your needs. Below is an example from my system on the Raspberry Pi. Remember to restart the NFS server when you have made changes to the file….
sudo service nfs-kernel-server restart
/etc/exports
# /etc/exports: the access control list for filesystems which may be exported
# to NFS clients. See exports(5).
#
# Example for NFSv2 and NFSv3:
# /srv/homes hostname1(rw,sync,no_subtree_check) hostname2(ro,sync,no_subtree_check)
#
# Example for NFSv4:
# /srv/nfs4 gss/krb5i(rw,sync,fsid=0,crossmnt,no_subtree_check)
# /srv/nfs4/homes gss/krb5i(rw,sync,no_subtree_check)
#
/srv/homes 192.168.1.9(rw,sync,no_subtree_check)
/home/erick 192.168.1.0/24(rw,all_squash,anonuid=1001,anongid=1004,no_subtree_check)
/ 192.168.1.9(rw,no_root_squash,anonuid=1001,anongid=1004,no_subtree_check)
My initial try at this was to semi follow an example and create a /srv/homes directory and export it to one machine at 192.168.1.9. rw = read write access, sync = change both folders on server and client to keep sync’d, no_subtree_check = keeps the machine from having to check consistency of file names, prevents problems if a file is open and the file is renamed.
Then I decided to export my own home directory to all of the machines on the LAN, using 192.168.1.0/24 which allows access from 192.168.1.1-192.168.1.255. This time I am using all_squash which maps all UID’s and GID’s to nobody and nogroup, then setting anonuid=1001, my UID on the Rasp Pi and anongid=1004 my group on the Rasp Pi, they will map over to the correct UID and GID for myself on the other machines. Therefore I have no problem with read write access as the same user on the other machine to the NFS drive.
The next line exports the entire file system to one machine, but has no_root_squash set, which allows the root to access and create files as root on the server. This is one to be careful with, I use it only when I need to mount the entire file system and usually I have to move things around or something as root anyways, but as always be cautious.
Restart Required
After modifying the /etc/exports file the NFS server needs to be restarted using…
service nfs-kernel-server restart
Client Machine
You have to install the common code for NFS on the client machine.
sudo apt-get install nfs-common
Mount Commands
My Home Directory on the Raspberry Pi
In this case I am mounting my home directory from the Pi under /home/erick-pi. I had to use the nolock option because I was getting an error without it, other than that it works fine.
Example of the mount command from the command line…
sudo mount -o nolock 192.168.1.17:/home/erick /home/erick-pi
Entire Root Directory
Occasionally I want to mount the entire file system of the Raspberry Pi at a location on one PC.
sudo mount -o nolock 192.168.1.17:/ /mnt/nfs/srv/
Mount Scripts
For situations that only apply occasionally, such as the above example of mounting the entire directory structure, I have created some scripts and placed them in the bin folder under my home folder and made them executable by using chmod +x filename. Then I can run them as needed by running a script with a filename that makes sense to me. Like the one for the code below is, rasp-pi-mount-root.sh.
#! /bin/bash
sudo mount -o nolock 192.168.1.17:/ /mnt/nfs/srv/
intr option
Note that the intr option for nfs mounts is a good one if the computer loses it’s connection with the server. With the Raspberry Pi this is never an issue for me but it is with other servers. It allows an interupt to stop NFS requests if the server goes down or the connection is lost. If intr is not used NFS will keep trying and the process will hang, requireing a reboot. I have had this occur mostly with servers that are on only part time. Most likely, I started the server and then put my computer on standby. When I start it and the server is off, NFS will hang looking for the mount points that have disappeared. This will hang not only the X window folders, but it will hang any command in a terminal that has to touch NFS, such as df -h which tries to look for something that is not there anymore.
Hard and Soft Options
The hard and soft options are like what they sound like. Soft mounts give up after a timeout and don’t keep trying to write or read from an NFS mount point if it flakes out or goes down. Soft mounts should only be used for read only mounts as data being written can be corrupted if a soft mount gives up where the hard mount will just keep trying until the mount comes back. If there is an issue with a write mount point flaking or dropping it is best to mount hard, the default and use the intr option.
Mounting on Startup
It is possible to set up the /etc/fstab file to mount the NFS drive on start up. It is as simple as adding the following line to the file…
192.168.1.17:/home/erick /home/erick-pi nolock 0 0
On my laptop this did not work, I remembered that Wireless LAN is handled at the user level and not on during bootup when mountpoints are handed out via the /etc/fstab file. So I got an error about the mount point not being found.
On my desktop running Lubuntu 14.04, connected via Ethernet cable, the line above did not work either but I modified it as follows and it worked. It might be that I left out the part with nfs, although I thought that the OS could tell it was a NFS mount from the format, oh well. I also decided to mount it right under my home folder on the desktop…
192.168.1.17:/home/erick /home/erick/erick-pi nfs auto,nolock 0 0
On the desktop once the drive is mounted it will stay mounted even if I reboot the Rasp Pi while the desktop PC is running.
Delayed Mount on Startup
To mount an NFS drive on a machine that has wireless, you have to mount it after it connects to the router and by then it is already running at the user level. You have to trick the system into waiting. There are multiple ways of doing this. I chose putting a line into to root crontab and used sleep 60 for the delay. After all most mounting has to be done as root anyways.
So I put a 60 second delay in before the mount command executes in the root crontab using the @reboot directive…
@reboot bash -c "sleep 60; mount -o nolock 192.168.1.17:/home/erick /home/erick-pi"
To edit the root crontab, simple do…
sudo nano crontab -e
To simply list what is in the root crontab, which is how I cut and pasted the code above, simply do…
sudo crontab -l
Using Names
It is entirely possible to use names instead of IP addresses when you mount NFS drives and even in the /etc/exports file. One caution, if DNS is down or flaky on you LAN, it could present a problem with reliably mounting drives. Therefore I recommend adding the server names to your /etc/hosts file. On my LAN I take it a step further, the servers are all set as static IP and my router has the ability to always hand out the same IP to a machine at a specific MAC address so I use that for laptops & etc that normally connect to the network. So in effect every normally used device has a static IP. Therefore I can put them all in /etc/hosts and I don’t even have to care about DNS on the LAN for the machines on it 99% of the time.
/etc/hosts comes with the top two entries, just add what you want to it. As you can see commenting them out works too. The erick-MS-6183 server is down, probably for good at this point!
127.0.0.1 localhost
127.0.1.1 erick-laptop
#192.168.1.11 erick-MS-6183
192.168.1.2 renee-pc
192.168.1.9 erick-laptop
192.168.1.10 ubuntuserver
192.168.1.17 raspberrypi
Gotcha
One time I was backing up my laptop to a laptop-backup directory under my home folder on the big file server. The problem was that I had my home folder on the big file server set as an NFS mount as a folder under my home folder on the laptop. It copied in circles until the harddrive filled up. Oh well, learned the hard way on that one! Be careful of NFS mounts and even symlinks to places when running backups.
NFS and Users
With users there is the notion of the name and then there is the numerical UID. NFS uses the numerical UID to map across machines. If you plan on using NFS on multiple machines, it pays to keep the UID’s lined up between them. For example, if you set up 2 Linux machines from scratch, there will be a user at UID 1000, that would be you, whatever you called it by name. The first user is at 1000. If you use NFS to mount a directory from one machine to another, no problem it all lines up. The user at UID 1000 is the same on both machines, permissions work out, files can be moved back and forth, no problems.
Resources
Used this one to get started with NFS…
https://help.ubuntu.com/community/SettingUpNFSHowTo
Helped to figure out the whole user and group ID mapping
NFS: Overview and Gotchas
exports(5) – Linux man page
Easy to follow, I hink I might have started with this one,
Setting up an NFS Server and Client on Debian Wheezy
I need to look at this one for a sanity check on the errors when I launch NFS server on Raspberrry Pi,
Problem with NFS network
