facebookarchive / nfusr

NFS client in userspace.

nfusr

nfusr is a userspace FUSE client for accessing NFSv3 servers based on libnfs.

Usage

NFS filesystems can be directly mounted with nfusr:

nfusr nfs://nfsserver.example.com/volume /mntpoint

(see libnfs documentation for the format of the nfs:// URI).

Alternatively one can mount using typical mount syntax (using the helper script installed in /sbin/mount.nfusr):

mount -t nfusr nfs://nfsserver.example.com/volume /mntpoint

which enables use of automount, etc.

In either case, FUSE mount options can be specified by adding a '-o' option, e.g.:

mount -t nfusr nfs://nfsserver.example.com/volume /mntpoint -o allow_other,max_read=128K

(see FUSE documentation for available FUSE mount options).

Note that NFSv4 is not supported in any way.

Building

To build from source:

./bootstrap
./configure
make
sudo make install

Debugging and output

By default nfusr deamonizes itself and messages are written to syslog. Alternatively output can be directed to a file using the --log-file option, e.g.

nfusr --log-file=/tmp/nfusr.log nfs://server/vol /mntpoint

In all instances the '--log-level' option controls the verbosity of the output using the levels defined in setlogmask(3) (i.e. --log-level=7 yields maximum verbosity).

The '-f' option forces nfusr to run in the foregound. In this case output is written to stdout.

The '-d' option enables libfuse debugging and implies -f.

Multi-server support

A novel feature of nfusr is support for multiple servers hosting the same volume, with round-robin load balancing and failover.

To mount a volume using multiple servers, specify multiple NFS URLs on the command line, and optionally specify the max_conn mount option. For example:

mount -t nfusr nfs://server1/vol,nfs://server2/vol /mntpoint

or:

nfusr nfs://server1/vol nfs://server2/vol nfs://server3/vol /mntpoint -o max_conn=2

When multiple servers are specified, nfusr will attempt to establish connections to max_conn servers (default 1) and round-robin requests amongst those connections.

If connection to a server is lost during a request, nfusr will attempt to connect to the next server and resend the request, thus failing over.

This feature comes with several caveats: first and foremost, the specified servers must be serving exactly the same volume, identical down to the structure of the NFS file handle (which is usually opaque). The only system against which this is known to reliably work at this time is Red Hat's Gluster distributed filesystem.

Other important caveats:

• fsync etc. should be considered broken in round-robin mode (currently the commit is sent to only one server, meaning writes sent to any others are not necessarily flushed despite a successful return).

• failover of non-idempotent requests can yield spurious errors. For example, if an unlink request is sent to a server and succeeds, but the connection is lost before the response is received by nfusr, the request will be replayed to another server. Since the file has actually been unlinked, the replayed request will now fail with ENOENT.

In short, multiple server mode is recommended only if you really, really know what you're doing.

License

nfusr is BSD-licensed. We also provide an additional patent grant.