NFS stands for Network File System and is just that, it is a good way to distribute filesystems, read only and read/write, while maintaining a degree of security and control assuming your network is enclosed and secure. NFS is primarily meant for use in a high bandwidth environment (i.e., a LAN) where security risks are not high, or the information being shared is not sensitive (i.e., a small trusted LAN behind a firewall exchanging CAD/CAM diagrams, or a large university lab using nfs to mount /usr/. If you need a high level of security, such as encrypting data between hosts, NFS is not the best choice. I personally use it at across my internal LAN (this machine has 2 interfaces, guess which one is heavily firewalled), to share file systems containing rpm's, this website, etc. Safer alternatives include SAMBA (free) and now IBM is porting AFS to Linux (costly, but AFS is a sweet piece of code).
NFS has a few rudimentary security controls. The first one would be firewalling; using NFS across a large, slow, public network like the Internet just isn't a good idea in any case, so firewall off port 2049, UDP. Since NFS runs as a set of daemons, TCP_WRAPPERS are of no use unless NFS is compiled to support them. The config file for NFS actually has quite a few directives, the bulk of which deal with user id and group id settings (map everyone to nobody, perhaps map all the engineering clients to 'engineer', etc, etc) but no real mechanisms for authentication (your client can claim to be UID 0, this is why root's id is squashed by default to nobody). NFS read-only exports are pretty safe, you only have to worry about the wrong people getting a look at your info (if it is sensitive) and or creating a denial of service attack (say you have a directory world readable/etc for sharing kernel source, and some gomer starts sucking down data like crazy...).
Writeable exports are a whole other ball game, and should be used with extreme caution, since the only 'authentication' is based on IP/hostname (both easily spoofable), and UID (you to can run Linux and be UID 0). Bounce a client down with a DOS attack, grab their IP, mount the writeable share and go to town. You say "but they'd have to know the IP and UID", packet sniffing is not rocket science folks, nor is 'showmount'.
So, how do we go about securing NFS? The first is to firewall it, especially if the machine is multi-homed, with an interface connected to a publicly accessible network (the Internet, the student lab, etc.). If you plan to run NFS over a publicly accessible network it better be read only, and you will be far better off with a different product then NFS.
The second and most interesting part is the /etc/exports file. This controls what you allow clients to do, and how they do it.
A sample exports file:
# # Allow a workstation to edit web content /www 10.0.0.11(rw,no_root_squash) # # Another share to allow a user to edit a web site /www/www.example.org 10.0.0.202(rw,no_root_squash) # # Public ftp directory /home/ftp *.example.org(ro,all_squash)
The structure of the exports file is pretty simple, directory
you wish to export, client (always use IPs, hostnames can
easily be faked), and any options. The client can be a single IP
(10.0.0.1), hostname (gomer.example.org), a subnet
(10.0.0.0/255.255.255.0), or a wildcard (*.example.org). Some of
the more interesting (and useful) directives for the exports file
secure - the nfs session must originate from a privileged port, i.e. root HAS to
be the one trying to mount the dir. This is useful if the server you are
exporting to is secured well.
ro - a good one, Read Only, enough said.
noaccess - used to cut off access, i.e. export /home/ but do a noaccess on /home/root
root_squash - squashes root's UID to the anonymous user UID/GID (usually 'nobody'), very useful if you are exporting dirs to servers with admins you do not 100% trust (root can almost always read any file.... HINT)
no_root_squash - useful if you want to go mucking about in exported dirs as root to fix things (like permissions on your www site)
squash_uids and squash_gids - squash certain UID(s) or GID(s) to the anonymous user, in Red Hat a good example would be 500-10000 (by default Red Hat starts adding users and groups at 500), allowing any users with lower UID's (i.e. special accounts) to access special things.
all_squash - a good one, all privileges are revoked basically and everyone is a guest.
anonuid and anongid - specifically set the UID / GID of the anonymous user (you might want something special like 'anonnfs').
The man exports page is actually quite good.
Beyond this there isn't much you can do to secure NFS apart from ripping it out and putting some other product in (like AFS, Coda, etc). NFS is relatively robust, almost every flavor of UNIX supports it, and it is usually easy to setup, work with and maintain. It's also 'old faithful', been around a long time. Just check "Practical Unix and Internet Security", they also state in bold not to use NFS if security is a real issue.
NFS should be restricted from the outside world, it runs on port 2049, udp, as well as using RPC which runs on port 111, udp/tcp, and makes use of mountd which runs on port 635, udp. Replace the 2049 with 111, and 635 udp and tcp to secure those services (again the best idea is a blanket rule to deny ports 1 to 1024, or better yet a default policy of denial).
ipfwadm -I -a accept -P udp -S 10.0.0.0/8 -D 0.0.0.0/0 2049 ipfwadm -I -a accept -P udp -S some.trusted.host -D 0.0.0.0/0 2049 ipfwadm -I -a deny -P udp -S 0.0.0.0/0 -D 0.0.0.0/0 2049
ipchains -A input -p udp -j ACCEPT -s 10.0.0.0/8 -d 0.0.0.0/0 2049 ipchains -A input -p udp -j ACCEPT -s some.trusted.host -d 0.0.0.0/0 2049 ipchains -A input -p udp -j DENY -s 0.0.0.0/0 -d 0.0.0.0/0 2049
Written by Kurt Seifried