Understand permission requirements for Docker Desktop on Mac

From version 4.18 and later, Docker Desktop for Mac provides greater control over functionality that's enabled during installation.

The first time Docker Desktop for Mac launches, it presents an installation window where you can choose to either use the default settings, which work for most developers and requires you to grant privileged access, or use advanced settings.

If you work in an environment with elevated security requirements, for instance where local administrative access is prohibited, then you can use the advanced settings to remove the need for granting privileged access. You can configure:

• The location of the Docker CLI tools either in the system or user directory
• The default Docker socket
• Privileged port mapping

Depending on which advanced settings you configure, you must enter your password to confirm.

You can change these configurations at a later date from the Advanced page in Settings.

Versions 4.15 to 4.17 of Docker Desktop for Mac don't require the privileged process to run permanently. Whenever elevated privileges are needed for a configuration, Docker Desktop prompts you with information on the task it needs to perform. Most configurations are applied once, subsequent runs don't prompt for privileged access anymore. The only time Docker Desktop may start the privileged process is for binding privileged ports that aren't allowed by default on the host OS.

Versions prior to 4.15 of Docker Desktop for Mac require root access to be granted on the first run. The first time that Docker Desktop launches you receive an admin prompt to grant permission for the installation of the com.docker.vmnetd privileged helper service. For subsequent runs, root privileges aren't required. Following the principle of least privilege, this approach allows root access to be used only for the operations for which it's absolutely necessary, while still being able to use Docker Desktop as an unprivileged user. All privileged operations are run using the privileged helper process com.docker.vmnetd.

With version 4.18 and later, you can choose whether to install symlinks either in /usr/local/bin or $HOME/.docker/bin during installation of Docker Desktop.

If /usr/local/bin is chosen, and this location is not writable by unprivileged users, Docker Desktop requires authorization to confirm this choice before the symlinks to Docker binaries are created in /usr/local/bin. If $HOME/.docker/bin is chosen, authorization is not required, but then you must manually add $HOME/.docker/bin to their PATH.

You are also given the option to enable the installation of the /var/run/docker.sock symlink. Creating this symlink ensures various Docker clients relying on the default Docker socket path work without additional changes.

As the /var/run is mounted as a tmpfs, its content is deleted on restart, symlink to the Docker socket included. To ensure the Docker socket exists after restart, Docker Desktop sets up a launchd startup task that creates the symlink by running ln -s -f /Users/<user>/.docker/run/docker.sock /var/run/docker.sock. This ensures the you aren't prompted on each startup to create the symlink. If you don't enable this option at installation, the symlink and the startup task is not created and you may have to explicitly set the DOCKER_HOST environment variable to /Users/<user>/.docker/run/docker.sock in the clients it is using. The Docker CLI relies on the current context to retrieve the socket path, the current context is set to desktop-linux on Docker Desktop startup.

For versions prior to 4.18, installing symlinks in /usr/local/bin is a privileged configuration Docker Desktop performs on the first startup. Docker Desktop checks if symlinks exists and takes the following actions:

• Creates the symlinks without the admin prompt if /usr/local/bin is writable by unprivileged users.
• Triggers an admin prompt for you to authorize the creation of symlinks in /usr/local/bin. If you authorizes this, symlinks to Docker binaries are created in /usr/local/bin. If you reject the prompt, are not willing to run configurations requiring elevated privileges, or don't have admin rights on your machine, Docker Desktop creates the symlinks in ~/.docker/bin and edits your shell profile to ensure this location is in your PATH. This requires all open shells to be reloaded. The rejection is recorded for future runs to avoid prompting you again. For any failure to ensure binaries are on your PATH, you may need to manually add to their PATH the /Applications/Docker.app/Contents/Resources/bin or use the full path to Docker binaries.

A particular case is the installation of the /var/run/docker.sock symlink. Creating this symlink ensures various Docker clients relying on the default Docker socket path work without additional changes. As the /var/run is mounted as a tmpfs, its content is deleted on restart, symlink to Docker socket included. To ensure the Docker socket exists after restart, Docker Desktop sets up a launchd startup task that creates a symlink by running ln -s -f /Users/<user>/.docker/run/docker.sock /var/run/docker.sock. This ensures that you are not prompted on each startup to create the symlink. If you reject the prompt, the symlink and the startup task are not created and you may have to explicitly set the DOCKER_HOST to /Users/<user>/.docker/run/docker.sock in the clients it is using. The Docker CLI relies on the current context to retrieve the socket path, the current context is set to desktop-linux on Docker Desktop startup.

With version 4.18 and later you can choose to enable privileged port mapping during installation, or from the Advanced page in Settings post-installation. Docker Desktop requires authorization to confirm this choice.

For versions below 4.18 , if you run a container that requires binding privileged ports, Docker Desktop first attempts to bind it directly as an unprivileged process. If the OS prevents this and it fails, Docker Desktop checks if the com.docker.vmnetd privileged helper process is running to bind the privileged port through it.

If the privileged helper process is not running, Docker Desktop prompts you for authorization to run it under launchd. This configures the privileged helper to run as in the versions of Docker Desktop prior to 4.15. However, the functionality provided by this privileged helper now only supports port binding and caching the Registry Access Management policy. If you decline the launch of the privileged helper process, binding the privileged port cannot be done and the Docker CLI returns an error:

$ docker run -p 127.0.0.1:80:80 docker/getting-started

docker: Error response from daemon: Ports are not available: exposing port
TCP 127.0.0.1:80 -> 0.0.0.0:0: failed to connect to /var/run/com.docker.vmnetd.sock:
is vmnetd running?: dial unix /var/run/com.docker.vmnetd.sock: connect: connection
refused.
ERRO[0003] error waiting for container: context canceled

Note

The command may fail with the same error if you take too long to authorize the prompt to start the helper process, as it may timeout.

With versions 4.18 and later, it is your responsibility to ensure that localhost is resolved to 127.0.0.1 and if Kubernetes is used, that kubernetes.docker.internal is resolved to 127.0.0.1.

On first run, Docker Desktop checks if localhost is resolved to 127.0.0.1. In case the resolution fails, it prompts you to allow adding the mapping to /etc/hosts. Similarly, when the Kubernetes cluster is installed, it checks that kubernetes.docker.internal is resolved to 127.0.0.1 and prompts you to do so.