Fakeroot feature


Fakeroot feature (or commonly referred as rootless mode) allows an unprivileged user to run a container as a “fake root” user by leveraging user namespace UID/GID mapping.


This feature requires a Linux kernel >= 3.8, but the recommended version is >= 3.18

A “fake root” user has almost the same administrative rights as root but only inside the container and the requested namespaces, which means that this user:

  • can set different user/group ownership for files or directories he owns
  • can change user/group identity with su/sudo commands
  • has full privileges inside the requested namespaces (network, ipc, uts)



A “fake root” user can’t access or modify files and directories for which he doesn’t have already access or rights on the host filesystem, so a “fake root” user won’t be able to access to host file like /etc/shadow or host /root directory.

Additionally, all files or directories created with “fake root” user are owned like root:root inside container and are owned like user:group outside of container. Let’s see the following example, in this case “user” is authorized to use the fakeroot feature and can use 65536 UIDs starting at 131072 (same thing for GIDs).

UID inside container UID outside container
0 (root) 1000 (user)
1 (daemon) 131072 (non-existent)
2 (bin) 131073 (non-existent)
65536 196607

Which means if “fake root” user creates a file a bin user in container, this file will be owned by 131073:131073 outside of container. The responsibility relies on the administrator to ensure that there is no overlap with the current user’s UID/GID on the system.


Restrictions are also applied for network, if singularity is executed without --net flag, the “fake root” user won’t be able to use ping or bind a container service on a port below than 1024.

With --net the “fake root” user has full privileges in this dedicated network, inside the container network he can bind on privileged ports below 1024, use ping, manage firewall rules, listen traffic … And everything done in this dedicated network won’t affect the host network.


Of course an unprivileged user could not map host ports below than 1024 by using: --network-args="portmap=80:80/tcp"


For unprivileged installation of Singularity or if allow setuid = no is set in singularity.conf users won’t be able to use fakeroot network.


Fakeroot looks at user mappings in /etc/subuid and /etc/subgid, so your username need to be listed in those files with a valid mapping (see admin-guide for details), if you can’t edit those files, ask to an administrator.

Then you could use it with --fakeroot or -f option, this option is available with singularity commands :

  • shell
  • exec
  • run
  • instance start
  • build


With fakeroot an unprivileged user can now build image from a definition file with few restrictions, some bootstrap methods requiring to create block devices (like /dev/null) may not always work correctly with “fake root”, Singularity uses seccomp filters to give programs the illusion that block devices creation worked, while it seems to work fine with yum and may work with other bootstrap methods, actually debootstrap is known to not work.


Build from a definition file:

singularity build --fakeroot /tmp/test.sif /tmp/test.def

Ping from container:

singularity exec --fakeroot --net docker://alpine ping -c1

HTTP server:

singularity run --fakeroot --net --network-args="portmap=8080:80/tcp" -w docker://nginx