Mounting and Unmounting Network Storage Devices

I'm using a small NAS storage server as network drive, which I mount the "Unix way" by NFS (Network File System), rather than using a Windows-compatible Samba client, which also would be supported by the device. That way, the mount point is accessible to all programs, not only the file manager or special clients that can use the smb:// protocol.

The command for mounting that I use under the directory name /media/volume1 is:

sudo mount -t nfs 10.0.0.20:/volume1 /media/volume1

Now, when suspending and disconnecting my computer from the network for traveling, I sometimes forget to unmount the NFS directory /media/volume1, which results in all programs freezing when trying to access that directory. Trying to unmount in that situation with

sudo umount /media/volume1

does not work because the device is busy. Even when no processes are accessing the directory, unmounting fails because the server is not reachable.

Is there an "official" way to unmount NFS volumes when the server is not reachable?

NFS was built to be reliable even in the case of network dropouts or temporary server disconnects; the client will wait for a long, sometimes even an unlimited time for the server to respond again. Even signals like TERM (15) and KILL (-9) won't terminate a process that waits for an answer from the NFS server.

Furthermore, you can even reboot the NFS server and the client still won't terminate. This is the desired behavior in NFS design.

Some mount options, like -o intr or -o timeo=n or -o retrans=n (see the NFS man page) change that behavior, but still you won't be able to unmount the NFS filesystem when the server is unreachable.

However, an easy trick allows you to work around the protocol: Just set the IP address of the former NFS server as the secondary IP address of your network card or loop back device first:

sudo ifconfig lo:1 10.0.0.20 netmask 255.255.255.255

For the client to detect the NFS server (it does not matter whether or not any directories are exported), start the port mapper and, if installed, the NFS kernel-based server:

sudo /etc/init.d/portmap start

or

sudo /etc/init.d/rpcbind start

then:

sudo /etc/init.d/nfs-kernel-server start

(The /etc/exports file should exist; otherwise, the init script may decide not to start the actual NFS server.) An rpc.mountd process should now be running, answering to mount and umount requests from the client.

The NFS client will now act as if the former NFS server is back online (even though there is really a new NFS server running on the client computer now), send a detach command on umount and flush all client accesses (provided no processes are still actively using the mount point):

sudo umount -l /media/volume1

I use the -l (lazy) option of umount here, which will allow running processes to continue accessing the directory but hide the mount point from new processes and remove the mtab entry, so it is not listed as a mounted volume anymore in /proc/mounts or in the output of mount or df with no parameters.

Another tip: To avoid too lengthy command lines for mounting NFS volumes or Samba shares, /etc/fstab entries can be helpful. Note that intr and nolock options for NFS allow you to cancel a read/write command by hitting Ctrl+C and avoids contacting the NFS locking process on the server side. Also, for NFS on Linux, the sync option (which is known to slow down write access to local media) can actually make NFS more responsive because it avoids buffering too much data in long writes, thus avoiding a traffic jam on wireless networks (Listing 1).

With these entries, the network storage directories can be mounted just by typing

mount /media/volume1

or

mount /media/backup

without sudo, as normal user.