If I have partition /dev/sda1 which is mounted on root /, and I have partition /dev/sdb1 which is mounted on /var, is there a way I can access original contents of /var on sda1 without first unmounting sdb1?


3 Answers 3


In Linux, you can use a bind mount to remount part of the file hierarchy somewhere else. So, for example, you could do:

# mkdir /mnt/bindroot
# mount --bind / /mnt/bindroot

At that point, /mnt/bindroot contain the contents of the root filesystem, but without other filesystems mounted on the various directories.

# ls /home
user1 lost+found

# ls /mnt/bindroot/home
<whatever was in /home before a filesystem was mounted over it>

For FreeBSD, you can do something similar with nullfs mounts -- see mount_nullfs.

  • right, "At that point" but after mount somewhere over / --bind can get cancelled, then just run it again ;-) Dec 1, 2021 at 13:25

On Linux, you can do that without a bind-mount and root privileges, by just having a process which either chdir's to or opens the underlying directory --before being hidden by the mount-- and then just lingers on; other processes could then access the directory via /proc/<pid>/cwd or /proc/<pid>/fd/<dir_fd>.

Example with chdir:

# mkdir dir; touch dir/file                   # create a sample dir and file
# (cd dir; while sleep 3600; do :; done) &    # start a bg process with its pwd being dir
[1] 3734
# mount -t tmpfs tmpfs dir                    # mount a tmpfs over dir which will hide its previous content
# ls dir
# ls /proc/3734/cwd                           # you can still access the old dir via /proc/<pid>/cwd 

Example with opening the directory:

# mkdir -p dir1; touch dir1/file
# exec 9<dir1
# mount -t tmpfs tmpfs dir1
# ls dir1
# ls /proc/self/fd/9

If you want to do that after the fact, with a directory which was already mounted over, you can create a private namespace and unmount that directory inside the namespace.

In this case, you can also access it from outside via /proc/<pid>/root/<path_to_dir>, without having to open it or chdir to it:

mkdir -p dir; touch dir/file
mount -t tmpfs tmpfs dir

unshare -m sh -c 'umount dir; while sleep 1000; do :; done' &
sleep .1
ls "/proc/$!/root/$PWD/dir"  # will show 'file'

You can read more about the extra magic of /proc/<pid>/root in the proc(5) manpage. Notice that you can keep the namespace of a process alive even after the process terminates by bind-mounting its /proc/<pid>/ns/mnt somewhere else; you can then re-enter it with nsenter(1).

  • 1
    In the context of the question, this would involve time travel, wouldn’t it? Oct 6, 2021 at 21:37
  • If that means that my answer is worthless and I should remove it: no way. It may help people coming from searches, which may try that mount-bind trick to access a directory which was already a mount point before being mounted over. Being aware of other options may help.
    – user313992
    Oct 7, 2021 at 20:50

Note: for those who may miss a comment.

sudo mount --bind original_location backup_link works, locations under oldlocation can be accessed after something is mounted over them and shadows.


  1. sudo mount --bind as I've found out need to be run after mount something original_location, not beforehand.
  2. As it is after shadowing mount, directory higher than one shadowed need to be mounted: sudo mount --bind parent_of_original_location backup_link
  • Discussed at length here
    – doneal24
    Dec 1, 2021 at 16:06

You must log in to answer this question.

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