Network File System (NFS)

NFS allows a system to share directories and files with others over a network. By using NFS, users and programs can access files on remote systems almost as if they were local files.

Some of the most notable benefits that NFS can provide are:

  • Local workstations use less disk space because commonly used data can be stored on a single machine and still remain accessible to others over the network.

  • There is no need for users to have separate home directories on every network machine. Home directories could be set up on the NFS server and made available throughout the network.

  • Storage devices such as floppy disks, CDROM drives, and USB Thumb drives can be used by other machines on the network. This may reduce the number of removable media drives throughout the network.

Installation

At a terminal prompt enter the following command to install the NFS Server:

sudo apt install nfs-kernel-server

To start the NFS server, you can run the following command at a terminal prompt:

sudo systemctl start nfs-kernel-server.service

Configuration

You can configure the directories to be exported by adding them to the /etc/exports file. For example:

/srv     *(ro,sync,subtree_check)
/home    *.hostname.com(rw,sync,no_subtree_check)
/scratch *(rw,async,no_subtree_check,no_root_squash)

Make sure any custom mount points you’re adding have been created (/srv and /home will already exist):

sudo mkdir /scratch

Apply the new config via:

sudo exportfs -a

You can replace * with one of the hostname formats. Make the hostname declaration as specific as possible so unwanted systems cannot access the NFS mount. Be aware that *.hostname.com will match foo.hostname.com but not foo.bar.my-domain.com.

The sync/async options control whether changes are gauranteed to be committed to stable storage before replying to requests. async thus gives a performance benefit but risks data loss or corruption. Even though sync is the default, it’s worth setting since exportfs will issue a warning if it’s left unspecified.

subtree_check and no_subtree_check enables or disables a security verification that subdirectories a client attempts to mount for an exported filesystem are ones they’re permitted to do so. This verification step has some performance implications for some use cases, such as home directories with frequent file renames. Read-only filesystems are more suitable to enable subtree_check on. Like with sync, exportfs will warn if it’s left unspecified.

There are a number of optional settings for NFS mounts for tuning performance, tightening security, or providing conveniences. These settings each have their own trade-offs so it is important to use them with care, only as needed for the particular use case. no_root_squash, for example, adds a convenience to allow root-owned files to be modified by any client system’s root user; in a multi-user environment where executables are allowed on a shared mount point, this could lead to security problems.

NFS Client Configuration

To enable NFS support on a client system, enter the following command at the terminal prompt:

sudo apt install nfs-common

Use the mount command to mount a shared NFS directory from another machine, by typing a command line similar to the following at a terminal prompt:

sudo mkdir /opt/example
sudo mount example.hostname.com:/srv /opt/example

Warning

The mount point directory /opt/example must exist. There should be no files or subdirectories in the /opt/example directory, else they will become inaccessible until the nfs filesystem is unmounted.

An alternate way to mount an NFS share from another machine is to add a line to the /etc/fstab file. The line must state the hostname of the NFS server, the directory on the server being exported, and the directory on the local machine where the NFS share is to be mounted.

The general syntax for the line in /etc/fstab file is as follows:

example.hostname.com:/srv /opt/example nfs rsize=8192,wsize=8192,timeo=14,intr

Advanced Configuration

NFS is comprised of several services, both on the server and the client. Each one of these services can have its own default configuration, and depending on the Ubuntu Server release you have installed, this configuration is done in different files, and with a different syntax.

Ubuntu Server 22.04 LTS (“jammy”)

All NFS related services read a single configuration file: /etc/nfs.conf. This is a INI-style config file, see the nfs.conf(5) manpage for details. Furthermore, there is a /etc/nfs.conf.d directory which can hold *.conf snippets that can override settings from previous snippets or from the nfs.conf main config file itself.

There is a new command-line tool called nfsconf(8) which can be used to query or even set configuration parameters in nfs.conf. In particular, it has a --dump parameter which will show the effective configuration including all changes done by /etc/nfs.conf.d/*.conf snippets.

For Ubuntu Server 20.04 LTS (“focal”) and earlier

Earlier Ubuntu releases use the traditional configuration mechanism for the NFS services via /etc/defaults/ configuration files. These are /etc/default/nfs-common and /etc/default/nfs/kernel-server, and are used basically to adjust the command-line options given to each daemon.

Each file has a small explanation about the available settings.

Warning

The NEED_* parameters have no effect on systemd-based installations, like Ubuntu 20.04 LTS (“focal”) and Ubuntu 18.04 LTS (“bionic”).
In those systems, to control whether a service should be running or not, use systemctl enable or systemctl disable, respectively.

Upgrading to Ubuntu 22.04 LTS (“jammy”)

The main change to the NFS packages in Ubuntu 22.04 LTS (“jammy”) is the configuration file. Instead of multiple files sourced by startup scripts from /etc/default/nfs-*, now there is one main configuration file in /etc/nfs.conf, with an INI-style syntax.
When upgrading to Ubuntu 22.04 LTS (“jammy”) from a release that still uses the /etc/defaults/nfs-* configuration files, the following will happen:

  • a default /etc/nfs.conf configuration file will be installed
  • if the /etc/default/nfs-* files have been modified, a conversion script will be run and it will create /etc/nfs.conf.d/local.conf with the local modifications.

If this conversion script fails, then the package installation will fail. This can happen if the /etc/default/nfs-* files have an option that the conversion script wasn’t prepared to handle, or a syntax error for example. In such cases, please file a bug using this link: https://bugs.launchpad.net/ubuntu/+source/nfs-utils/+filebug

You can run the conversion tool manually to gather more information about the error: it’s in /usr/share/nfs-common/nfsconvert.py and must be run as root.

If all goes well, as it should in most cases, the system will have /etc/nfs.conf with the defaults, and /etc/nfs.conf.d/local.conf with the changes. You can merge these two together manually, and then delete local.conf, or leave it as is. Just keep in mind that /etc/nfs.conf is not the whole story: always inspect /etc/nfs.conf.d as well, as it may contain files overriding the defaults.

You can always run nfsconf --dump to check the final settings, as it merges together all configuration files and shows the resulting non-default settings.

Restarting NFS services

Since NFS is comprised of several individual services, it can be difficult to determine what to restart after a certain configuration change.

The tables below summarize all available services, which “meta” service they are linked to, and which configuration file each service uses.

Service nfs-utils.service nfs-server.service config file (22.04) config file (< 22.04) /etc/default/nfs-*
nfs-blkmap PartOf nfs.conf
nfs-mountd BindsTo nfs.conf nfs-kernel-server
nfsdcld
nfs-idmapd BindsTo nfs.conf, idmapd.conf idmapd.conf
rpc-gssd PartOf nfs.conf
rpc-statd PartOf nfs.conf nfs-common
rpc-svcgssd PartOf BindsTo nfs.conf nfs-kernel-server

For example, systemctl restart nfs-server.service will restart nfs-mountd, nfs-idmapd and rpc-svcgssd (if running). On the other hand, restarting nfs-utils.service will restart nfs-blkmap, rpc-gssd, rpc-statd and rpc-svcgssd.

Of course, each service can still be individually restarted with the usual systemctl restart <service>.

The nfs.systemd(7) manpage has more details on the several systemd units available with the NFS packages.

NFS with Kerberos

Kerberos with NFS adds an extra layer of security on top of NFS. It can be just a stronger authentication mechanism, or it can also be used to sign and encrypt the NFS traffic.

This section will assume you already have setup a Kerberos server, with a running KDC and admin services. Setting that up is explained elsewhere in the Ubuntu Server Guide.

NFS server (using kerberos)

The NFS server will have the usual nfs-kernel-server package and its dependencies, but we will also have to install kerberos packages. The kerberos packages are not strictly necessary, as the necessary keys can be copied over from the KDC, but it makes things much easier.

For this example, we will use:

  • .vms DNS domain
  • VMS Kerberos realm
  • j-nfs-server.vms for the NFS server
  • j-nfs-client.vms for the NFS client
  • ubuntu/admin principal has admin privileges on the KDC

Adjust these names according to your setup.

First, install the krb5-user package:

sudo apt install krb5-user

Then, with an admin principal, let’s create a key for the NFS server:

$ sudo kadmin -p ubuntu/admin -q "addprinc -randkey nfs/j-nfs-server.vms"

And extract the key into the local keytab:

$ sudo kadmin -p ubuntu/admin -q "ktadd nfs/j-nfs-server.vms"
Authenticating as principal ubuntu/admin with password.
Password for ubuntu/admin@VMS:
Entry for principal nfs/j-nfs-server.vms with kvno 2, encryption type aes256-cts-hmac-sha1-96 added to keytab FILE:/etc/krb5.keytab.
Entry for principal nfs/j-nfs-server.vms with kvno 2, encryption type aes128-cts-hmac-sha1-96 added to keytab FILE:/etc/krb5.keytab.

Confirm the key is available:

$ sudo klist -k
Keytab name: FILE:/etc/krb5.keytab
KVNO Principal
---- --------------------------------------------------------------------------
   2 nfs/j-nfs-server.vms@VMS
   2 nfs/j-nfs-server.vms@VMS

Now install the NFS server:

$ sudo apt install nfs-kernel-server

This will already automatically start the kerberos-related nfs services, because of the presence of /etc/krb5.keytab.

Now populate /etc/exports, restricting the exports to krb5 authentication. For example, exporting /storage using krb5p:

/storage *(rw,sync,no_subtree_check,sec=krb5p)

Refresh the exports:

$ sudo exportfs -rav
exporting *:/storage

The security options are explained in the exports(5) manpage, but generally they are:

  • krb5: use kerberos for authentication only (non-auth traffic is in clear text)
  • krb5i: use kerberos for authentication and integrity checks (non-auth traffic is in clear text)
  • krb5p: use kerberos for authentication, integrity and privacy protection (non-auth traffic is encrypted)

NFS client (using kerberos)

The NFS client has a similar set of steps. First we will prepare the client’s keytab, so that when we install the NFS client package it will start the extra kerberos services automatically just by detecting the presence of the keytab:

sudo apt install krb5-user

To allow the root user to mount NFS shares via kerberos without a password, we have to create a host key for the NFS client:

sudo kadmin -p ubuntu/admin -q "addprinc -randkey host/j-nfs-client.vms"

And extract it:

$ sudo kadmin -p ubuntu/admin -q "ktadd host/j-nfs-client.vms"

Now install the NFS client package:

$ sudo apt install nfs-common

And you should be able to do your first NFS kerberos mount:

$ sudo mount j-nfs-server:/storage /mnt

If you are using a machine credential, then the above mount will work without having a kerberos ticket, i.e., klist will show no tickets:

# mount j-nfs-server:/storage /mnt
# ls -l /mnt/*
-rw-r--r-- 1 root root 0 Apr  5 14:50 /mnt/hello-from-nfs-server.txt
# klist
klist: No credentials cache found (filename: /tmp/krb5cc_0)

Notice the above was done with root. Let’s try accessing that existing mount with the ubuntu user, without acquiring a kerberos ticket:

# sudo -u ubuntu -i
$ ls -l /mnt/*
ls: cannot access '/mnt/*': Permission denied

The ubuntu user will only be able to access that mount if they have a kerberos ticket:

$ kinit
Password for ubuntu@VMS: 
$ ls -l /mnt/*
-rw-r--r-- 1 root root 0 Apr  5 14:50 /mnt/hello-from-nfs-server.txt

And now we have not only the TGT, but also a ticket for the NFS service:

$ klist
Ticket cache: FILE:/tmp/krb5cc_1000
Default principal: ubuntu@VMS

Valid starting     Expires            Service principal
04/05/22 17:48:50  04/06/22 03:48:50  krbtgt/VMS@VMS
        renew until 04/06/22 17:48:48
04/05/22 17:48:52  04/06/22 03:48:50  nfs/j-nfs-server.vms@
        renew until 04/06/22 17:48:48
        Ticket server: nfs/j-nfs-server.vms@VMS

One drawback of using a machine credential for mounts done by the root user is that you need a persistent secret (the /etc/krb5.keytab file) in the filesystem. Some sites may not allow such a persistent secret to be stored in the filesystem. An alternative is to use rpc.gssds -n option. From rpc.gssd(8):

  • -n: when specified, UID 0 is forced to obtain user credentials which are used instead of the local system’s machine credentials.

When this option is enabled and rpc.gssd restarted, then even the root user will need to obtain a kerberos ticket to perform an NFS kerberos mount.

Warning

Note that this prevents automatic NFS mounts via /etc/fstab, unless a kerberos ticket is obtained before.

In Ubuntu 22.04 LTS (“jammy”), this option is controlled in /etc/nfs.conf in the [gssd] section:

[gssd]
use-machine-creds=0

In older Ubuntu releases, the command line options for the rpc.gssd daemon are not exposed in /etc/default/nfs-common, therefore a systemd override file needs to be created. You can either run:

$ sudo systemctl edit rpc-gssd.service

And paste the following into the editor that will open:

[Service]
ExecStart=
ExecStart=/usr/sbin/rpc.gssd $GSSDARGS -n

Or manually create the file /etc/systemd/system/rpc-gssd.service.d/override.conf and any needed directories up to it, with the contents above.

After you restart the service with systemctl restart rpc-gssd.service, the root user won’t be able to mount the NFS kerberos share without obtaining a ticket first.

References

Linux NFS wiki
Linux NFS faq

Ubuntu Wiki NFS Howto
Ubuntu Wiki NFSv4 Howto

1 Like

Documentation clean up.
config section of service-nfs shows “noexec” as an option in one of the examples.

but in 20.04 it’s not recognized:

root@spinningrust:/etc# exportfs -rav
exportfs: /etc/exports:2: unknown keyword “noexec”

root@spinningrust:/etc# man exports.5 | grep noexec
root@spinningrust:/etc#

Thanks, I removed it. It’s indeed invalid for /etc/exports: it’s an /etc/fstab option.

The heading “NFSv4 with Kerberos” should IMHO just say “NFS with Kerberos”, because NFSv3 and NFSv4 work the same way with Kerberos. I’ve been using Kerberized NFSv3 every day for more than 15 years, and v3 still has advantages, such as no id-mapping (and all the things that can go wrong there), as well as fewer bugs in many implementations (e.g., macOS long had a far more robust NFSv3 than NFSv4 implementation).

Kerberos isn’t part of NFSv4, but is implemented in the RPC layer underneath (as RPCSEC_GSS), which is used by both NFSv3 and NFSv4.

Thanks for your comment. You are correct, and I removed the “v4” part from that header.