Network File System (NFS)

The Network File System is the protocol of choice to share files over an internal Local Area Network. Depending on your needs, you may also want to use Samba or the SSH Filesystem additionally or instead.

Normally an OpenWRT host acting as an NFS server will have external storage attached (e.g. USB). Assuming clients access the NFS server from the LAN zone, OpenWRT's default configuration should not need any changes to the firewall to allow client access.

1. usb-installing obtain basic support for USB.
2. usb-drives obtain support for USB storage and mount local filesystem
3. In your firewall (or see netfilter) open port 111 tcp and udp, and 32777-32780 tcp and udp from your LAN. This should be true of a default OpenWRT configuration. If needed, an appropriate set of firewall rules allowing NFS looks like the following:

uci add firewall rule
uci set firewall.@rule[-1].name='Allow-NFS-RPC'
uci set firewall.@rule[-1].src='lan'
uci set firewall.@rule[-1].proto='tcp udp'
uci set firewall.@rule[-1].dest_port='111'
uci set firewall.@rule[-1].target='ACCEPT'
 
uci add firewall rule
uci set firewall.@rule[-1].name='Allow-NFS'
uci set firewall.@rule[-1].src='lan'
uci set firewall.@rule[-1].proto='tcp udp'
uci set firewall.@rule[-1].dest_port='2049'
uci set firewall.@rule[-1].target='ACCEPT'
 
uci add firewall rule
uci set firewall.@rule[-1].name='Allow-NFS-Lock'
uci set firewall.@rule[-1].src='lan'
uci set firewall.@rule[-1].proto='tcp udp'
uci set firewall.@rule[-1].dest_port='32777:32780'
uci set firewall.@rule[-1].target='ACCEPT'
 
uci commit firewall
service firewall restart

The portmap service uses port 111 on both TCP and UDP, nfsd standard are ports between 32777 and 32780 on both TCP and UDP.
It might be necessary to populate /etc/hosts.allow to whitelist NFS clients, insert an entry like portmap: 192.168.1.* then.

NOTE: You do not need all packages: nfs-kernel-server should suffice.

NOTE: You may need to install kmod-loop manually to fix “mount: can't setup loop device: No such file or directory” errors. (see https://dev.openwrt.org/ticket/11541)

GNU/Linux distribution come with pre-installed support for this. So does MacOS X since version 10.2. In case your distribution is missing support, you need to install the client software. ArchLinux offers a Wikipage: https://wiki.archlinux.org/index.php/Samba#Configuration.

For Windows download Microsoft Windows Services for UNIX from Microsoft.

You can also use JFtp as an alternative client.

opkg

opkg update
opkg install nfs-kernel-server
vi /etc/exports
. /etc/init.d/portmap start
. /etc/init.d/portmap enable
. /etc/init.d/nfsd start
. /etc/init.d/nfsd enable
netstat -a
$IPT -A INPUT -j ACCEPT -i eth0.1 -s $NET_LAN -p tcp --dport 111 #------------------- portmap
$IPT -A INPUT -j ACCEPT -i eth0.1 -s $NET_LAN -p udp --dport 111 #------------------- portmap
$IPT -A INPUT -j ACCEPT -i eth0.1 -s $NET_LAN -p tcp --dport 32777:32780 #----------- nfsd
$IPT -A INPUT -j ACCEPT -i eth0.1 -s $NET_LAN -p udp --dport 32777:32780 #----------- nfsd
$IPT -t raw -A INPUT -i $IF_LAN -s $NET_LAN -p tcp --dport 32777:32780 -j CT --notrack #-- don't track nfs
$IPT -t raw -A INPUT -i $IF_LAN -s $NET_LAN -p udp --dport 32777:32780 -j CT --notrack #-- don't track nfs
$IPT -t raw -A OUTPUT -o $IF_LAN -d $NET_LAN -p tcp --dport 32777:32780 -j CT --notrack #- don't track nfs
$IPT -t raw -A OUTPUT -o $IF_LAN -d $NET_LAN -p udp --dport 32777:32780 -j CT --notrack #- don't track nfs

We have a typical client «» server configuration.

Use the file /etc/exports to configure your shares. NFSv4 export paths don't work the way they did in NFSv3. NFSv4 has a global root directory (configured as fsid=0) and all exported directories are children to it. So what would have been nfs-server:/export/users on NFSv3 is nfs-server:/users on NFSv4, because /export is the root directory. Example:

See exports(5) for configuration semantics. A single asterisk matches all IP addresses (allowing anonymous access).

If you set up pivot-root or pivot-overlay, u;se the path on /overlay/ partition, else you cannot export mounted fs.

Assuming the daemons are already running, use the command exportfs -ar to reload and apply changes on the fly.

Since opkg-installed init.d scripts are not enabled nor started by default, you need to do this manually. Same procedure as with most (all?) OpenWrt packages: The first commands will will start them right now, the third and fourth will create the symlinks /etc/rc.d/S??portmap and /etc/rc.d/S??nfsd so they get started on boot:

root@OpenWrt:~# /etc/init.d/portmap start
root@OpenWrt:~# /etc/init.d/nfsd start
root@OpenWrt:~# /etc/init.d/portmap enable
root@OpenWrt:~# /etc/init.d/nfsd enable

After start (and after a reboot) verify with top or ps whether the services are running.
The following entries should appear in the process list:

/usr/sbin/portmap -f
/usr/sbin/rpc.mountd -p 32780 -F
various nfsd instances
nfsiod
nfsd4_callbacks

Use the netstat -l command to see whether portmap is listening on port 111 for both tcp and udp. The nfsd process may use varying ports.

Mount manually:

sudo mount 192.168.1.254:/sda1 /home/sandra/nfs_share

Or mount permanently with entries in the /etc/fstab on each client PC:

Check the manual for mount and take a particular look at the options. Choose wisely.

With portmap running on your OpenWrt-Machine you can use rpcinfo -p 192.168.1.254 on clients side to see open ports. The clients must be in /etc/hosts.allow.

Java client: JFtp. TODO

Java client: . TODO

If the loopback device support is missing, an error like “Cannot register service: RPC: Timed out” may appear. Installing the kmod-loop package should solve this issue.

If portmap is not registering mountd daemon, you may need to install rpcbind package:

# install
opkg install rpcbind # (if it is in your repos)

# disable portmap
/etc/init.d/portmap disable

# enable
/etc/init.d/rpcbind enable
/etc/init.d/nfsd enable

# start
/etc/init.d/rpcbind start
/etc/init.d/nfsd start

If opkg can't find it you probably have to build it yourself Build a package.

NB: Always restart services after rpcbind, this order is important!

Maybe you want to compare the read and write performance with that of your samba setup. Post it in the forum or right here:

Do not expect to much. When i started with my RouterStationPro i got 7 MB/s writing to and 10 MB/s reading from it.

Now:

Server:

• echo 4 > /proc/fs/nfsd/threads
• /etc/exports: /mnt/bla 192.168.1.0/255.255.255.0(rw,all_squash,insecure,async,no_subtree_check)
• Check out /proc/fs/nfsd/max_block_size if client reports other block size then what you have specified. It can be changed if the server is not running, but nfsd mounted. Now you can echo the right value to it. - Updated by fathom

Client: mount options: udp,async,rsize=32768,wsize=32768,intr,noatime

i get: writing: 15,3MB/s and reading: 17,4MB/s

I have TL-WR1043ND from TP-Link by default it has a throughput of 4,5mb/sec, after tweaking the block size, and the mount options it delivers 9,5 mb/sec. Cheers! (Still on wifi it has a speed of 6 mb/s)

Since netfilter will track every connection, if you use MASQUERADING for example, you could disable con-tracking for data connections:

$IPT -t raw -A PREROUTING -i $IF_LAN -s $NET_LAN -p tcp --dport 32777:32780 -j CT --notrack #---------- don't track nfs
$IPT -t raw -A PREROUTING -i $IF_LAN -s $NET_LAN -p udp --dport 32777:32780 -j CT --notrack #---------- don't track nfs
$IPT -t raw -A OUTPUT -o $IF_LAN -d $NET_LAN -p tcp --sport 32777:32780 -j CT --notrack #---------- don't track nfs
$IPT -t raw -A OUTPUT -o $IF_LAN -d $NET_LAN -p udp --sport 32777:32780 -j CT --notrack #---------- don't track nfs