Skip to main content

A basic user tool to execute simple docker containers in batch or interactive systems without root privileges

Project description

Build Status


udocker is a basic user tool to execute simple docker containers in user space without requiring root privileges. Enables download and execution of docker containers by non-privileged users in Linux systems where docker is not available. It can be used to pull and execute docker containers in Linux batch systems and interactive clusters that are managed by other entities such as grid infrastructures or externally managed batch or interactive systems.

udocker does not require any type of privileges nor the deployment of services by system administrators. It can be downloaded and executed entirely by the end user.

udocker is a wrapper around several tools to mimic a subset of the docker capabilities including pulling images and running containers with minimal functionality.

How does it work

udocker is a simple tool written in Python, it has a minimal set of dependencies so that can be executed in a wide range of Linux systems.

udocker does not make use of docker nor requires its presence.

udocker "executes" the containers by simply providing a chroot like environment over the extracted container. The current implementation supports different methods to mimic chroot enabling execution of containers without requiring privileges under a chroot like environment. udocker transparently supports several methods to execute the containers using tools and libraries such as:

  • PRoot
  • Fakechroot
  • runC
  • Singularity


  • Provides a docker like command line interface
  • Supports a subset of docker commands: search, pull, import, export, load, create and run
  • Understands docker container metadata
  • Can be deployed by the end-user
  • Does not require privileges for installation
  • Does not require privileges for execution
  • Does not require compilation, just transfer the Python script and run
  • Encapsulates several execution methods
  • Includes the required tools already compiled to work across systems
  • Tested with GPGPU and MPI applications
  • Runs both on new and older Linux distributions including: CentOS 6, CentOS 7, Ubuntu 14, Ubuntu 16, Ubunto 18, Fedora, etc


See the Installation manual


  search <repo/image:tag>       :Search dockerhub for container images
  pull <repo/image:tag>         :Pull container image from dockerhub
  images                        :List container images
  create <repo/image:tag>       :Create container from a pulled image
  ps                            :List created containers
  rm  <container>               :Delete container
  run <container>               :Execute container
  inspect <container>           :Low level information on container
  name <container_id> <name>    :Give name to container
  rmname <name>                 :Delete name from container

  rmi <repo/image:tag>          :Delete image
  rm <container-id>             :Delete container
  import <tar> <repo/image:tag> :Import tar file (exported by docker)
  import - <repo/image:tag>     :Import from stdin (exported by docker)
  load -i <exported-image>      :Load image from file (saved by docker)
  load                          :Load image from stdin (saved by docker)
  export -o <tar> <container>   :Export container rootfs to file
  export - <container>          :Export container rootfs to stdin
  inspect <repo/image:tag>      :Return low level information on image
  verify <repo/image:tag>       :Verify a pulled image
  clone <container>             :duplicate container

  protect <repo/image:tag>      :Protect repository
  unprotect <repo/image:tag>    :Unprotect repository
  protect <container>           :Protect container
  unprotect <container>         :Unprotect container

  mkrepo <topdir>               :Create repository in another location
  setup                         :Change container execution settings
  login                         :Login into docker repository
  logout                        :Logout from docker repository

  help                          :This help
  run --help                    :Command specific help

Options common to all commands must appear before the command:
  -D                          :Debug
  --repo=<directory>          :Use repository at directory


Some examples of usage:

Search container images in dockerhub.

udocker search  fedora
udocker search  ubuntu
udocker search  indigodatacloud

Pull from dockerhub and list the pulled images.

udocker pull  fedora:25
udocker pull  busybox
udocker pull  iscampos/openqcd
udocker images

Pull from a registry other than dockerhub.

udocker pull --registry=  rhel7
udocker create --name=rh7 rhel7
udocker run rh7

Create the container from a pulled image and run it.

udocker create --name=myfed  fedora:25
udocker run  myfed  cat /etc/redhat-release

Run mounting the host /home/u457 into the container directory /home/cuser. Notice that you can "mount" any host directory inside the container, this is not a real mount but the directories will be visible inside the container.

udocker run -v /home/u457:/home/cuser -w /home/user myfed  /bin/bash
udocker run -v /var -v /proc -v /sys -v /tmp  myfed  /bin/bash

Put a script in your host /tmp and execute it in the container.

udocker run  -v /tmp  myfed  /bin/bash -c 'cd /tmp; ./'

Run mounting the host /var, /proc, /sys and /tmp in the same container directories. Notice that the content of these container directories will be obfuscated.

udocker run -v /var -v /proc -v /sys -v /tmp  myfed  /bin/bash

Install software inside the container.

udocker run  --user=root myfed  yum install -y firefox pulseaudio gnash-plugin

Run as some user. The usernames should exist in the container.

udocker run --user 1000:1001  myfed  /bin/id
udocker run --user root   myfed  /bin/id
udocker run --user jorge  myfed  /bin/id

Running Firefox.

./udocker run --bindhome --hostauth --hostenv \
   -v /sys -v /proc -v /var/run -v /dev --user=jorge --dri myfed  firefox

Change execution engine mode from PRoot to Fakechroot and run.

./udocker setup  --execmode=F4  myfed

./udocker run --bindhome --hostauth --hostenv \
   -v /sys -v /proc -v /var/run -v /dev --user=jorge --dri myfed  firefox

Change execution engine mode to accelerated PRoot.

./udocker setup  --execmode=P1  myfed

Change execution engine to runC.

./udocker setup  --execmode=R1  myfed

Change execution engine to Singularity. Requires the availability of Singularity in the host system.

./udocker setup  --execmode=S1  myfed


Since root privileges are not involved any operation that really requires such privileges will not be possible. The following are examples of operations that are not possible:

  • accessing host protected devices and files
  • listening on TCP/IP privileged ports (range below 1024)
  • mount file-systems
  • the su command will not work
  • change the system time
  • changing routing tables, firewall rules, or network interfaces

If the containers require such capabilities then docker should be used instead.

The current implementation is limited to the pulling of docker images and its execution. The actual containers should be built using docker and dockerfiles.

udocker does not provide all the docker features, and is not intended as a docker replacement.

Debugging inside of udocker with the PRoot engine will not work due to the way PRoot implements the chroot environment

udocker is mainly oriented at providing a run-time environment for containers execution in user space.

udocker is particularly suited to run user applications encapsulated in docker containers.


Because of the limitations described in the previous section udocker does not offer isolation features such as the ones offered by docker. If the containers content is not trusted then these containers should not be executed with udocker as they will run inside the user environment.

The containers data will be unpacked and stored in the user home directory or other location of choice. Therefore the containers data will be subjected to the same filesystem protections as other files owned by the user. If the containers have sensitive information the files and directories should be adequately protected by the user.

udocker does not require privileges and runs under the identity of the user invoking it.

Users can downloaded udocker and execute it without requiring system administrators intervention.

udocker via PRoot offers the emulation of the root user. This emulation mimics a real root user (e.g getuid will return 0). This is just an emulation no root privileges are involved. This feature makes possible the execution of some tools that do not require actual privileges but which refuse to work if the username or id are not root or 0. This enables for instance software installation using rpm, yum or dnf inside the container.

Due to the lack of isolation udocker must not be run by privileged users.

Other limitations

Notice that when using execution engines other than PRoot (Pn modes) the created containers cannot be moved across hosts. In this case convert back to a Pn mode before transfer.

The accelerated mode of PRoot (mode P1) may exhibit failures in Linux kernels above 4.0 with some applications due to kernel changes and upstream issues in this case use mode P2 or any of the other modes.

The runC mode requires a recent kernel with user namespaces enabled.

The singularity mode requires the availability of Singularity in the host system.


The full documentation is available at:


See: Contributing


When citing udocker please use the following:

  • Jorge Gomes, Emanuele Bagnaschi, Isabel Campos, Mario David, Luís Alves, João Martins, João Pina, Alvaro López-García, Pablo Orviz, Enabling rootless Linux Containers in multi-user environments: The udocker tool, Computer Physics Communications, Available online 6 June 2018, ISSN 0010-4655,


udocker (1.1.3)

  • Support for nvidia drivers on ubuntu
    • closes: #162
  • Installation improvements
    • closes: #166
  • Fix issue on Fn mode symlink convertion
    • partially addresses: #160

udocker (1.1.2)

  • Improve parsing of quotes in the command line
    • closes: #98
  • Fix version command to exit with 0
    • closes: #107
  • Add kill-on-exit to proot on Pn modes
  • Improve download of udocker utils
  • Handle authentication headers when pulling
    • closes: #110
  • Handle of redirects when pulling
  • Fix registries table
  • Support search
  • Fix auth header when no standard Docker registry is used
  • Add registry detection on image name
  • Add --version option
  • Force python2 as interpreter
    • closes: #131
  • Fix handling of volumes in metadata
  • Handle empty metadata
  • Fix http proxy functionality
    • closes: #115
  • Ignore --no-trunc and --all in the images command
    • closes: #108
  • Implement verification of layers in manifest
  • Add --nvidia to support GPUs and related drivers
  • Send download messages to stderr
  • Enable override of curl executable
  • Fix building on CentOS 6
    • closes: #157
  • Mitigation for upstream limitation in runC without tty
    • closes: #132
  • Fix detection of executable with symlinks in container
    • closes: #118
  • Updated runC to v1.0.0-rc5
  • Experimental support for Alpine in Fn modes
  • Improve pathname translation in Fn modes for mounted dirs
    • partially addresses: #160

udocker (1.1.1)

  • New execution engine using singularity
  • Updated documentation with OpenMPI information and examples
  • Additional unit tests
  • Redirect messages to stderr
  • Improved parsing of quotes in the command line
    • closes: #87
  • Allow override of the HOME environment variable
  • Allow override of at the container level
  • Automatic selection of from container info
  • Improve automatic install
  • Enable resetting prefix paths in Fn modes in remote hosts
  • Do not set AF_UNIX_PATH in Fn modes when the host /tmp is a volume
  • Export containers in both docker and udocker format
  • Import containers docker and udocker format
  • Load, import and export to/from stdin/stdout
  • Clone existing containers
  • Support for TCP/IP port remap in execution modes Pn
  • Fix run with basenames failing
    • closes: #89
  • Allow run as root flag
    • closes: #91

udocker (1.1.0)

  • Support image names prefixed by registry similarly to docker
  • Add execution engine selection logic
  • Add fr execution engine based on shared library interception
  • Add rc execution engine based on rootless namespaces
  • Improve proot tmp files cleanup on non ext filesystems
  • Improve search returning empty on Docker repositories
  • Improve runC execution portability
  • Add environment variable UDOCKER_KEYSTORE
    • closes: #75
  • Prevent creation of .udocker when UDOCKER_KEYSTORE is used
    • closes: #75

udocker (1.0.4)

  • Documentation fixes

udocker (1.0.3)

  • Support for import Docker containers in newer metadata structure
  • Improve the command line parsing
  • Improve temporary file handling and removal
  • Support for additional execution engines to be provided in the future
  • Improved parsing of entrypoint and cmd metadata
    • closes: #53
  • Increase name alias length
    • closes: #52
  • Add support for change dir into volume directories
    • closes: #51
  • Fix deletion of files upon container import
    • closes: #50
  • Fix exporting of host environment variables to the containers
    • closes: #48
  • Change misleading behavior of import tarball from move to copy
    • closes: #44
  • Fix validation of volumes specification
    • closes: #43

udocker (1.0.2)

  • Improve download on repositories that fail authentication on /v2
  • Improve run verification of binaries with recursive symbolic links
  • Improve accelerated seccomp on kernels >= 4.8.0
    • closes: #40

udocker (1.0.1)

  • Minor bugfixes
  • Executable name changed from to udocker
  • Added support for login into docker repositories
  • Added support for private repositories
  • Added support for listing of v2 repositories catalog
  • Added checksum verification for sha256 layers
  • Improved download handling for v1 and v2 repositories
  • Improved installation tarball structure
  • Insecure flag fixed
  • Address seccomp change introduced on kernels >= 4.8.0
  • Utilities for packaging
  • Improved verbose levels, messaging and output
    • closes: #24, #23
  • Fully implement support for registry selection --registry parameter
    • closes: #29
  • Provide support for private repositories e.g. gitlab registries
    • closes: #30
  • Provide --insecure command line parameter for SSL requests
    • closes: #31

udocker (1.0.0)

  • Initial version

Project details

Release history Release notifications

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for udocker-stimela, version 1.1.4
Filename, size File type Python version Upload date Hashes
Filename, size udocker-stimela-1.1.4.tar.gz (153.2 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page