Create a tool sandboxcontainer that will allow an administrator to easily create a LXC Container to run the an application server, with libvirt and SELinux locking it down.
This change allows an administrator to run multiple services on the same machine each service in a secure Linux Container. sandboxcontainer takes an executable and sets up all of the mountpoints and libvirt container information to run the application within the container. Then libvirt will be able to launch the container with a SELinux context that will prevent the containers process from interacting with other process on the system including other containers. This could allow an administrator to run multiple web servers simultaneously each with their own data and sharing the system data, even running as root, but prevent them from breaking the host machine or other processes on the system.
- Name: Daniel J Walsh
- Email: firstname.lastname@example.org
- Targeted release: [Fedora 17]
- Last updated: Mon Nov 7 2011
- Percentage of completion: 20%
libvirt as of Fedora 16 has the ability to run processes within a container, the problem with this is the processes can run as root and can easily break out of the container. Also it is fairly difficult to setup one of these containers to run a particular service, for example apache.
We want to make the processes of setting up these containers easier and to use SELinux and containers together to lock down the processes running within the container, meaning you could run multiple customers on a host and prevent them from attacking each other or the host.
Benefit to Fedora
Linux Containers is a low level way of doing virtualization, and allows you to run multiple copies of the same service at the same time on a system. It has some advantages over full virtualization as it does not have to wait for an entire system to boot, it can use less memory and can use the base OS, in a read/only manner.
This change effects
- SELinux Policy writers
- New policy has to be written to define what a confined application container is allowed to do
- Kernel and SELinux toolchain changes
- One problem we have with the way the kernel currently labels files, is that MCS portion of the label is applied based off the process level not the level of the directory that the file was create in. This means when a process running as unconfined_t:s0 creates a file in a directory labeled svirt_lxc_file_t:s0:c1,c2, the file will be created as svirt_lxc_file_t:s0 and the container will not be able to use the file, until the administrator changed the label.
This feature requires that the kernel support new policy construct to allow the policy writer to specify whether the MCS Label is inherited from the parent directory MCS Labelor from the creating process MCS Label.
- Has to be able to launch the container with a new SELinux context for containers,
- potentially libvirt will have to mount a container file image with a random MCS label
- New tool needs to be created which will look at the content related to an executable, by examining the rpm database, then optionally create an image file or a directory on disk and populate the image/directory with the default content to be used for the container. It will also create the XML database used by libvirt and actually create the virtual machine definition. Finally it will create a default "init" which libvirt will execute when it starts the linux container.
How To Test
Users should be able to easily establish containers and be able to run multiple containers simultaneously.