Forcefully Unmount a Busy Filesystem in Linux

When you try to unmount a filesystem, Linux sometimes refuses with a “device is busy” error. This safety mechanism prevents data loss when processes are actively using the filesystem. But there are legitimate cases where you need to force an unmount: a dead NFS server, a failed USB device, or a hung mount that won’t recover.

Understanding the “device is busy” error

Before forcing anything, it’s worth knowing what’s holding the filesystem open. Use lsof to list open files on a specific mount:

lsof /mnt/point

Or use fuser to find processes using a directory:

fuser -m /mnt/point

This tells you which PIDs are keeping the filesystem busy. You might be able to simply kill those processes and unmount cleanly.

Lazy unmount with -l flag

The safest forced unmount is a lazy unmount using the -l flag:

umount -l /mnt/point

This detaches the filesystem from the namespace immediately, so the mount point becomes inaccessible. However, the actual cleanup happens later when all processes release their references. It’s especially useful for NFS mounts and removable media where you need immediate removal but aren’t concerned about lingering cleanup.

Force unmount with -f flag

For filesystems that are completely unresponsive (like a dead NFS server), use the -f flag:

umount -f /mnt/point

This tells the kernel to forcefully unmount the filesystem without waiting for processes to close files. The behavior depends on the filesystem type — some filesystems ignore this flag entirely.

Combining flags

You can combine both flags for maximum force:

umount -fl /mnt/point

This does a lazy unmount and also sends a force signal. It’s useful when a mount is completely hung.

Unmounting from a different namespace

If the mount is in another container or namespace, you need to enter that namespace first. For a container:

nsenter -t <pid> -m umount /mnt/point

Or for systemd containers:

machinectl shell container-name /bin/bash
umount /mnt/point

Killing processes holding the mount

If you identify specific processes preventing unmount, you can terminate them first:

fuser -km /mnt/point

The -k flag kills all processes accessing that mount. The -m flag specifically targets the mount. Be careful — this will terminate whatever is using the filesystem.

Dealing with rootfs and system mounts

If you’re trying to unmount the root filesystem or a critical system mount, you’re fighting the kernel itself. In these cases:

• Boot into a live USB or recovery mode
• Use remount,ro to make the filesystem read-only instead
• Consider using mount --bind to redirect the mount point

mount -o remount,ro /

A read-only root filesystem is safer than a hung one and won’t prevent system shutdown.

NFS-specific issues

For dead NFS mounts, lazy unmount is your best friend:

umount -l /mnt/nfs

If that doesn’t work, you may need to restart the NFS client service:

systemctl restart nfs-client.target

Then retry the unmount. If the NFS server is truly unreachable, a lazy unmount with -l is your only option.

Preventing busy filesystems in the future

Set mount options that prevent this issue:

mount -o soft,timeo=10,retrans=3 server:/export /mnt/point

The soft option tells NFS to fail quickly instead of hanging indefinitely. Lower timeout values (timeo) and retry counts (retrans) prevent the kernel from holding onto the mount too long.

Use -t flag with umount to specify filesystem type if the kernel has trouble detecting it, particularly with older or exotic filesystem types.