116

I'm running Ubuntu 11.10 - setting up NFS to share a directory among many other servers. Which ports are required to be opened on the firewall?

1
  • 11
    It depends on the version of the protocol you intent to use. NFS 4 only require 2049 while older versions require more.
    – lzap
    Sep 11, 2015 at 9:26

8 Answers 8

154
$ rpcinfo -p | grep nfs

Port 111 (TCP and UDP) and 2049 (TCP and UDP) for the NFS server.

There are also ports for Cluster and client status (Port 1110 TCP for the former, and 1110 UDP for the latter) as well as a port for the NFS lock manager (Port 4045 TCP and UDP). Only you can determine which ports you need to allow depending on which services are needed cross-gateway.

8
  • 8
    I did not know about rpcinfo, that is quite useful. I didn't see the port 111 with the grep nfs, but I left off the grep to learn that 111 is for portmapper. Also good to know! (and like you mentioned, necessary)
    – kenny
    Apr 5, 2012 at 21:12
  • 1
    @KennyYounger rpcinfo is a useful too. If this answered your question, don't forget to mark it as the answer for the benefit of future viewers.
    – Wesley
    Apr 5, 2012 at 21:17
  • 7
    You need mountd open when you first mount the filesystem. It runs on a dynamic port, so it won't always be the same. I put a link to a guide in my answer. Apr 5, 2012 at 21:23
  • 1
    Thanks @bonsaiviking. Critical info for this process. I upvoted your answer!
    – kenny
    Apr 6, 2012 at 0:25
  • 2
    Could you give a brief description of what those services ("cluster and client status", "NFS lock manager") do and when they might be needed? If those services are blocked by a firewall, will NFS servers and clients degrade gracefully, or just lock waiting for a response? Apr 9, 2015 at 0:41
42

In addition to 111 for portmapper and 2049 for nfs, you will need to allow the mountd port and possibly rquotad, lockd, and statd, all of which can be dynamic. This excellent NFS security guide recommends changing your startup scripts and kernel module configs to force them to use static ports.

In addition to the guide above, which has a section on firewalls, see my answer to another question about hardening NFS.

3
  • 11
    We really do prefer content, not pointers to content. A précis of the content with a link is ok too and preferred to a link.
    – user9517
    Apr 5, 2012 at 21:31
  • 3
    Not sure about 2012 but the recommendation in links seems to be grossly outdated. NFS over TCP is preferred for variety of reason and NFS over UDP can cause silent data corruption on fast link due to protocol limitations. Security guide has no mention of Kerberos/GSS etc. Dec 26, 2016 at 16:46
  • With respect to mountd, life is a lot easier if you reconfigure to ensure the mountd port is fixed. On debian systems, edit the RPCMOUNTDOPTS value to something like RPCMOUNTDOPTS="--port 34567" and then restart with sysctl --system and /etc/init.d/nfs-kernel-server restart
    – user45793
    Jan 17, 2019 at 19:53
32

I found useful directions for my problem on this page, but there was no easy to follow recipe. So here's my recipe.

TL;DR - need to allow both nfs ports (111, 2049) and mountd port after fixing it.

Instructions:


Setting up a fixed port for mountd

gksudo gedit /etc/default/nfs-kernel-server
  • comment out this line: RPCMOUNTDOPTS=--manage-gids
  • add this instead: RPCMOUNTDOPTS="--port 33333"

Or any other port number.

now try to reset nfs using:

sudo service nfs-kernel-server restart

And test if it helped using:

rpcinfo -p | grep "tcp.*mountd"

For me it wasn't enough, but a full restart fixed the issue.

(credit)


Setting up the firewall (ufw)

(1) delete old rules, do this manually or reset if this is the only use for the firewall:

# WARNING: Don't copy & paste this if you don't understand what it does:
# sudo ufw reset
# sudo ufw enable

(2) add nfs & mountd ports

sudo ufw allow in from 10.0.0.1/20 to any port 111 
sudo ufw allow in from 10.0.0.1/20 to any port 2049
sudo ufw allow in from 10.0.0.1/20 to any port 33333

(Change to your local IP's or to "any" instead of 10.0.0.1/20)

That's all there's to it.

6
  • 3
    You don't need to restart the entire box. A simple sudo service nfs-config restart before restarting the nfs-kernel-server will do just fine.
    – showp1984
    Feb 27, 2017 at 23:29
  • Or if your system uses systemctl, it's systemctl restart nfs-kernel-server.service.
    – fbicknel
    Oct 7, 2019 at 20:27
  • You can also do it all in one swell foop: sudo nfs allow proto tcp from 10.0.0.1/20 to any port 111,2049,33333 followed by (ok, two swell foops) sudo nfs allow proto udp from 10.0.0.1/20 to any port 111,2049,33333. Note port 33333 is per the above example. You may use another port as the answer indicates.
    – fbicknel
    Oct 7, 2019 at 20:30
  • @fbicknel I think you can do it in one "swell foop" if you just remove the proto tcp part on the first command :)
    – jchook
    Apr 24, 2020 at 16:40
  • 1
    @jchook That looks factual to me!
    – fbicknel
    Apr 25, 2020 at 17:14
10

This will give a list of all ports used by all NFS-related program:

rpcinfo -p | awk '{print $3" "$4}' | sort -k2n | uniq
2

With FERM one can use Backticks to get the ports from rpcinfo, for example:

Server:

proto tcp {saddr ($CLIENT) {
  dport (`rpcinfo -p | perl -e 'while(<>){/\s+\d+\s+\d\s+(?:tcp)\s+(\d+)/ and $ports{$1}=1}; $, = " "; print sort(keys(%ports)),"\n"'`) ACCEPT; # NFS
}}
proto udp {saddr ($CLIENT) {
  dport (`rpcinfo -p | perl -e 'while(<>){/\s+\d+\s+\d\s+(?:udp)\s+(\d+)/ and $ports{$1}=1}; $, = " "; print sort(keys(%ports)),"\n"'`) ACCEPT; # NFS
}}

Client:

proto udp {saddr ($SERVER) {ACCEPT;}}  # NFS

(If you're only going to use the TCP then you need only the proto tcp part).

2

To mount Synology to Ubuntu 18.04 system, I had to enable ports 111,892,2049

When mounting, here is what I see (NFS 4 is not enabled on my Synology):

root@ub18ovh# mount -a -vv
mount.nfs: trying text-based options 'vers=4.2,addr=5.6.7.8,clientaddr=1.2.3.4'
mount.nfs: mount(2): Protocol not supported
mount.nfs: trying text-based options 'vers=4.1,addr=5.6.7.8,clientaddr=1.2.3.4'
mount.nfs: mount(2): Protocol not supported
mount.nfs: trying text-based options 'vers=4.0,addr=5.6.7.8,clientaddr=1.2.3.4'
mount.nfs: mount(2): Protocol not supported
mount.nfs: trying text-based options 'addr=5.6.7.8'
mount.nfs: prog 100003, trying vers=3, prot=6
mount.nfs: trying 5.6.7.8 prog 100003 vers 3 prot TCP port 2049
mount.nfs: prog 100005, trying vers=3, prot=17
mount.nfs: trying 5.6.7.8 prog 100005 vers 3 prot UDP port 892
successfully mounted
1

For the records, I had to add permissions for ports 111, 2049 AND 1048 for a configuration where an NFS share is exported by a Windows 2008 R2 server and the clients are Ubuntu 12.04.4.

I hope this helps someone.

2
  • 2
    It might help to explain why you needed port 1048 and how you determined that.
    – HBruijn
    Dec 16, 2015 at 10:48
  • 3
    Frankly, I don't have a clue why I needed to add a permission for 1048, but adding that solved my problem. I just wanted to share if that can save time someone else in the future. I am sorry for not being able to answer the question.
    – specstr
    Dec 20, 2015 at 10:33
1

if you use csf firewall and nfs does not mount you most likely miss the open ports used by nlockmgr, find them by typing

rpcinfo -p

Next edit /etc/sysctl.conf to LOCK the ports on these numbers (example port) and add these 2 lines. Then restart portmap, nfs-server.

fs.nfs.nlm_udpport=38073 

fs.nfs.nlm_tcpport=38747

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .