Terminology

A cgroup is a collection of processes that are bound to a set of limits or parameters defined via the cgroup filesystem.

A subsystem is a kernel component that modifies the behavior of the processes in a cgroup. Various subsystems have been implemented, making it possible to do things such as limiting the amount of CPU time and memory available to a cgroup, accounting for the CPU time used by a cgroup, and freezing and resuming execution of the processes in a cgroup. Subsystems are sometimes also known as resource controllers (or simply, controllers).

The cgroups for a subsystem are arranged in a hierarchy. This hierarchy is defined by creating, removing, and renaming subdirectories within the cgroup filesystem. At each level of the hierarchy, attributes (e.g., limits) can be defined; these attributes may govern or propagate to child cgroups and their descendants in the hierarchy.

Cgroups version 1 and version 2

The initial release of the cgroups implementation was in Linux 2.6.24. Over time, various cgroup subsystems have been added to allow the management of various types of resources. However, the development of these subsystems was largely uncoordinated, with the result that many inconsistencies arose between subsystems and management of the cgroup hierarchies became rather complex. (A longer description of these problems can be found in the kernel source file Documentation/cgroup−v2.txt.)

Because of the problems with the initial cgroups implementation (cgroups version 1), starting in Linux 3.10, work began on a new, orthogonal implementation to remedy these problems. Initially marked experimental, and hidden behind the −o __DEVEL__sane_behavior mount option, the new version (cgroups version 2) was eventually made official with the release of Linux 4.5. Differences between the two versions are described in the text below.

Although cgroups v2 is intended as a replacement for cgroups v1, the older system continues to exist (and for compatibility reasons is unlikely to be removed). Currently, cgroups v2 implements only a subset of the controllers available in cgroups v1. The two systems are implemented so that both v1 controllers and v2 controllers can be mounted on the same system. Thus, for example, it is possible to use those controllers that are supported under version 2, while also using version 1 controllers where version 2 does not yet support those controllers.

Tasks versus processes

In cgroups v1, a distinction is drawn between processes and tasks. In this view, a process can consist of multiple tasks (more commonly called threads, from a user-space perspective). In cgroups v1, it is possible independently manipulate the cgroup memberships of the tasks in a process. Because this ability caused certain problems, the ability to independently manipulate the cgroup memberships of the tasks in a process has been removed in cgroups v2. Cgroups v2 allows manipulation of cgroup membership only for processes (which has the effect of changing the cgroup membership of all tasks in the process).

Mounting

To be available, a given cgroup subsystem must be compiled into the kernel. Since they are exposed through a virtual filesystem, subsystems must be mounted before they can be controlled. The usual place for this is under /sys/fs/cgroup. If all the desired subsystems can be comounted, then one can do so with the following command:

mount −t cgroup −o all cgroup /sys/fs/cgroup

(One can achieve the same result by omitting −o all, since it is the default if subsystems are explicitly specified.)

If multiple, separately mounted subsystems are desired, then this is usually done in per-subsystem subdirectories. This requires first mounting a tmpfs under /sys/fs/cgroup so that subdirectories can be created. For instance, one could mount cpu, memory, and devices cgroups as follows:

Comounting subsystems has the effect that a task is in the same cgroup for all comounted subsystems. Separately mounting subsystems allows a task to be in cgroup /foo1 for one subsystem while being in /foo2/foo3 for another.

Introspection

The list of subsystems compiled into the kernel can be seen in the file /proc/cgroups. The file /proc/pid/cgroup lists the task's current cgroup membership for each mounted hierarchy.

Creating cgroups and moving tasks

The system begins with a single root cgroup (per hierarchy), '/', which all tasks belong to. A new cgroup is created by creating a directory in the cgroup filesystem:

mkdir /sys/fs/cgroup/cpu/cg1

This creates a new empty cgroup. Tasks may be moved to this cgroup by writing their PIDs into the cgroup's cgroup.procs or tasks (deprecated) file:

echo $$ > /sys/fs/cgroup/cpu/cg1/cgroup.procs

The same file can be read to obtain a list of the processes currently in cg1. By using the cgroup.procs file instead of the tasks file, all tasks in the thread group are moved into the new cgroup at once.

On fork(2), the new child is created as a member of the parent's cgroup, leading to implicit grouping of process hierarchies.

Note

In the upcoming unified hierarchy, a new restriction is imposed such that tasks may exist only in leaf cgroups. For instance, if cgroup /cg1/cg2 exists, then a task may exist in /cg1/cg2, but not in /cg1. This is to avoid the current ambiguity in the delegation of resources between tasks in /cg1 and its child cgroups. The recommended workaround is to create a subdirectory called leaf for any non-leaf cgroup which should contain tasks, and make sure not to create child cgroups of it. In the above example, tasks which previously would have gone into /cg1 would now go into /cg1/leaf. This has the advantage of making explicit the relationship between tasks in /cg1/leaf and /cg1's other children.

Removing cgroups

To remove a cgroup, it must first have no child cgroups and contain no tasks. So long as that is the case, one can simply remove the corresponding directory pathname. Note that files in a cgroup directory cannot and need not be removed.

A special file in each cgroup hierarchy, release_agent, can be used to register a program to handle cgroups which become newly empty. The program will be called each time a cgroup marked for autoremove becomes empty and childless. The cgroup path will be provided as the first command-line argument. The cgroup must be marked as eligible for autoremove by writing '1' into its notify_on_release file; this value is inherited by newly created child cgroups.

A new feature in cgroups v2 is the cgroup.populated file. This reads 0 if there are no tasks in the cgroup or its descendants, and 1 otherwise. It can be watched for changes using inotify(7). This allows user-space applications to efficiently watch cgroups for autoremove conditions.

Cgroups version 2

In cgroups v2, all mounted controllers reside in a single unified hierarchy. While (different) controllers may be simultaneously mounted under the v1 and v2 hierarchies, it is not possible to mount the same controller simultaneously under both the v1 and the v2 hierarchies.

The new behaviors in cgroups v2 are summarized below:

Cgroups version 1 subsystems

Each of the cgroups version 1 subsystems is governed by a kernel configuration option (listed below). Additionally, the availability of the cgroups feature is governed by the CONFIG_CGROUPS kernel configuration option.

/proc files