blob: 638bf17ff86965a561b5cc258451ea53bd752f1a [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001 CGROUPS
2 -------
3
Li Zefan45ce80f2009-01-15 13:50:59 -08004Written by Paul Menage <menage@google.com> based on
5Documentation/cgroups/cpusets.txt
Paul Menageddbcc7e2007-10-18 23:39:30 -07006
7Original copyright statements from cpusets.txt:
8Portions Copyright (C) 2004 BULL SA.
9Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
10Modified by Paul Jackson <pj@sgi.com>
11Modified by Christoph Lameter <clameter@sgi.com>
12
13CONTENTS:
14=========
15
161. Control Groups
17 1.1 What are cgroups ?
18 1.2 Why are cgroups needed ?
19 1.3 How are cgroups implemented ?
20 1.4 What does notify_on_release do ?
Daniel Lezcano97978e62010-10-27 15:33:35 -070021 1.5 What does clone_children do ?
22 1.6 How do I use cgroups ?
Paul Menageddbcc7e2007-10-18 23:39:30 -0700232. Usage Examples and Syntax
24 2.1 Basic Usage
25 2.2 Attaching processes
Kirill A. Shutemov8ca712e2010-03-10 15:22:10 -080026 2.3 Mounting hierarchies by name
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -080027 2.4 Notification API
Paul Menageddbcc7e2007-10-18 23:39:30 -0700283. Kernel API
29 3.1 Overview
30 3.2 Synchronization
31 3.3 Subsystem API
Aristeu Rozanski19ec2562012-09-11 16:28:10 -0400324. Extended attributes usage
335. Questions
Paul Menageddbcc7e2007-10-18 23:39:30 -070034
351. Control Groups
Li Zefand19e0582008-02-23 15:24:08 -080036=================
Paul Menageddbcc7e2007-10-18 23:39:30 -070037
381.1 What are cgroups ?
39----------------------
40
41Control Groups provide a mechanism for aggregating/partitioning sets of
42tasks, and all their future children, into hierarchical groups with
43specialized behaviour.
44
45Definitions:
46
47A *cgroup* associates a set of tasks with a set of parameters for one
48or more subsystems.
49
50A *subsystem* is a module that makes use of the task grouping
51facilities provided by cgroups to treat groups of tasks in
52particular ways. A subsystem is typically a "resource controller" that
53schedules a resource or applies per-cgroup limits, but it may be
54anything that wants to act on a group of processes, e.g. a
55virtualization subsystem.
56
57A *hierarchy* is a set of cgroups arranged in a tree, such that
58every task in the system is in exactly one of the cgroups in the
59hierarchy, and a set of subsystems; each subsystem has system-specific
60state attached to each cgroup in the hierarchy. Each hierarchy has
61an instance of the cgroup virtual filesystem associated with it.
62
Chris Samuelcaa790b2009-01-17 00:01:18 +110063At any one time there may be multiple active hierarchies of task
Paul Menageddbcc7e2007-10-18 23:39:30 -070064cgroups. Each hierarchy is a partition of all tasks in the system.
65
Michael Kerrisk83b061f2012-09-11 13:20:20 +020066User-level code may create and destroy cgroups by name in an
Paul Menageddbcc7e2007-10-18 23:39:30 -070067instance of the cgroup virtual file system, specify and query to
Michael Kerrisk83b061f2012-09-11 13:20:20 +020068which cgroup a task is assigned, and list the task PIDs assigned to
Paul Menageddbcc7e2007-10-18 23:39:30 -070069a cgroup. Those creations and assignments only affect the hierarchy
70associated with that instance of the cgroup file system.
71
72On their own, the only use for cgroups is for simple job
73tracking. The intention is that other subsystems hook into the generic
74cgroup support to provide new attributes for cgroups, such as
75accounting/limiting the resources which processes in a cgroup can
Michael Kerrisk83b061f2012-09-11 13:20:20 +020076access. For example, cpusets (see Documentation/cgroups/cpusets.txt) allow
Paul Menageddbcc7e2007-10-18 23:39:30 -070077you to associate a set of CPUs and a set of memory nodes with the
78tasks in each cgroup.
79
801.2 Why are cgroups needed ?
81----------------------------
82
83There are multiple efforts to provide process aggregations in the
Michael Kerrisk83b061f2012-09-11 13:20:20 +020084Linux kernel, mainly for resource-tracking purposes. Such efforts
Paul Menageddbcc7e2007-10-18 23:39:30 -070085include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server
86namespaces. These all require the basic notion of a
87grouping/partitioning of processes, with newly forked processes ending
Michael Kerrisk83b061f2012-09-11 13:20:20 +020088up in the same group (cgroup) as their parent process.
Paul Menageddbcc7e2007-10-18 23:39:30 -070089
90The kernel cgroup patch provides the minimum essential kernel
91mechanisms required to efficiently implement such groups. It has
92minimal impact on the system fast paths, and provides hooks for
93specific subsystems such as cpusets to provide additional behaviour as
94desired.
95
96Multiple hierarchy support is provided to allow for situations where
97the division of tasks into cgroups is distinctly different for
98different subsystems - having parallel hierarchies allows each
99hierarchy to be a natural division of tasks, without having to handle
100complex combinations of tasks that would be present if several
101unrelated subsystems needed to be forced into the same tree of
102cgroups.
103
104At one extreme, each resource controller or subsystem could be in a
105separate hierarchy; at the other extreme, all subsystems
106would be attached to the same hierarchy.
107
108As an example of a scenario (originally proposed by vatsa@in.ibm.com)
109that can benefit from multiple hierarchies, consider a large
110university server with various users - students, professors, system
111tasks etc. The resource planning for this server could be along the
112following lines:
113
Geunsik Lim6ad85232011-04-04 15:10:45 -0700114 CPU : "Top cpuset"
Paul Menageddbcc7e2007-10-18 23:39:30 -0700115 / \
116 CPUSet1 CPUSet2
Geunsik Lim6ad85232011-04-04 15:10:45 -0700117 | |
118 (Professors) (Students)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700119
120 In addition (system tasks) are attached to topcpuset (so
121 that they can run anywhere) with a limit of 20%
122
Geunsik Lim6ad85232011-04-04 15:10:45 -0700123 Memory : Professors (50%), Students (30%), system (20%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700124
Geunsik Lim6ad85232011-04-04 15:10:45 -0700125 Disk : Professors (50%), Students (30%), system (20%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700126
127 Network : WWW browsing (20%), Network File System (60%), others (20%)
128 / \
Geunsik Lim6ad85232011-04-04 15:10:45 -0700129 Professors (15%) students (5%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700130
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200131Browsers like Firefox/Lynx go into the WWW network class, while (k)nfsd goes
132into the NFS network class.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700133
Chris Samuelcaa790b2009-01-17 00:01:18 +1100134At the same time Firefox/Lynx will share an appropriate CPU/Memory class
Paul Menageddbcc7e2007-10-18 23:39:30 -0700135depending on who launched it (prof/student).
136
137With the ability to classify tasks differently for different resources
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200138(by putting those resource subsystems in different hierarchies),
Paul Menageddbcc7e2007-10-18 23:39:30 -0700139the admin can easily set up a script which receives exec notifications
140and depending on who is launching the browser he can
141
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700142 # echo browser_pid > /sys/fs/cgroup/<restype>/<userclass>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700143
144With only a single hierarchy, he now would potentially have to create
145a separate cgroup for every browser launched and associate it with
Jörg Sommer67de0162011-06-15 13:00:47 -0700146appropriate network and other resource class. This may lead to
Paul Menageddbcc7e2007-10-18 23:39:30 -0700147proliferation of such cgroups.
148
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200149Also let's say that the administrator would like to give enhanced network
Paul Menageddbcc7e2007-10-18 23:39:30 -0700150access temporarily to a student's browser (since it is night and the user
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200151wants to do online gaming :)) OR give one of the student's simulation
152apps enhanced CPU power.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700153
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200154With ability to write PIDs directly to resource classes, it's just a
155matter of:
Paul Menageddbcc7e2007-10-18 23:39:30 -0700156
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700157 # echo pid > /sys/fs/cgroup/network/<new_class>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700158 (after some time)
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700159 # echo pid > /sys/fs/cgroup/network/<orig_class>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700160
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200161Without this ability, the administrator would have to split the cgroup into
Paul Menageddbcc7e2007-10-18 23:39:30 -0700162multiple separate ones and then associate the new cgroups with the
163new resource classes.
164
165
166
1671.3 How are cgroups implemented ?
168---------------------------------
169
170Control Groups extends the kernel as follows:
171
172 - Each task in the system has a reference-counted pointer to a
173 css_set.
174
175 - A css_set contains a set of reference-counted pointers to
176 cgroup_subsys_state objects, one for each cgroup subsystem
177 registered in the system. There is no direct link from a task to
178 the cgroup of which it's a member in each hierarchy, but this
179 can be determined by following pointers through the
180 cgroup_subsys_state objects. This is because accessing the
181 subsystem state is something that's expected to happen frequently
182 and in performance-critical code, whereas operations that require a
183 task's actual cgroup assignments (in particular, moving between
Paul Menage817929e2007-10-18 23:39:36 -0700184 cgroups) are less common. A linked list runs through the cg_list
185 field of each task_struct using the css_set, anchored at
186 css_set->tasks.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700187
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200188 - A cgroup hierarchy filesystem can be mounted for browsing and
Paul Menageddbcc7e2007-10-18 23:39:30 -0700189 manipulation from user space.
190
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200191 - You can list all the tasks (by PID) attached to any cgroup.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700192
193The implementation of cgroups requires a few, simple hooks
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200194into the rest of the kernel, none in performance-critical paths:
Paul Menageddbcc7e2007-10-18 23:39:30 -0700195
196 - in init/main.c, to initialize the root cgroups and initial
197 css_set at system boot.
198
199 - in fork and exit, to attach and detach a task from its css_set.
200
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200201In addition, a new file system of type "cgroup" may be mounted, to
Paul Menageddbcc7e2007-10-18 23:39:30 -0700202enable browsing and modifying the cgroups presently known to the
203kernel. When mounting a cgroup hierarchy, you may specify a
204comma-separated list of subsystems to mount as the filesystem mount
205options. By default, mounting the cgroup filesystem attempts to
206mount a hierarchy containing all registered subsystems.
207
208If an active hierarchy with exactly the same set of subsystems already
209exists, it will be reused for the new mount. If no existing hierarchy
210matches, and any of the requested subsystems are in use in an existing
211hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy
212is activated, associated with the requested subsystems.
213
Tejun Heo26d5bbe2013-04-12 10:29:04 -0700214It's not currently possible to bind a new subsystem to an active
215cgroup hierarchy, or to unbind a subsystem from an active cgroup
216hierarchy. This may be possible in future, but is fraught with nasty
217error-recovery issues.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700218
219When a cgroup filesystem is unmounted, if there are any
220child cgroups created below the top-level cgroup, that hierarchy
221will remain active even though unmounted; if there are no
222child cgroups then the hierarchy will be deactivated.
223
224No new system calls are added for cgroups - all support for
225querying and modifying cgroups is via this cgroup file system.
226
227Each task under /proc has an added file named 'cgroup' displaying,
228for each active hierarchy, the subsystem names and the cgroup name
229as the path relative to the root of the cgroup file system.
230
231Each cgroup is represented by a directory in the cgroup file system
232containing the following files describing that cgroup:
233
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200234 - tasks: list of tasks (by PID) attached to that cgroup. This list
235 is not guaranteed to be sorted. Writing a thread ID into this file
Paul Menage7823da32009-10-07 16:32:26 -0700236 moves the thread into this cgroup.
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200237 - cgroup.procs: list of thread group IDs in the cgroup. This list is
238 not guaranteed to be sorted or free of duplicate TGIDs, and userspace
Paul Menage7823da32009-10-07 16:32:26 -0700239 should sort/uniquify the list if this property is required.
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200240 Writing a thread group ID into this file moves all threads in that
Ben Blum74a11662011-05-26 16:25:20 -0700241 group into this cgroup.
Li Zefand19e0582008-02-23 15:24:08 -0800242 - notify_on_release flag: run the release agent on exit?
243 - release_agent: the path to use for release notifications (this file
244 exists in the top cgroup only)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700245
246Other subsystems such as cpusets may add additional files in each
Li Zefand19e0582008-02-23 15:24:08 -0800247cgroup dir.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700248
249New cgroups are created using the mkdir system call or shell
250command. The properties of a cgroup, such as its flags, are
251modified by writing to the appropriate file in that cgroups
252directory, as listed above.
253
254The named hierarchical structure of nested cgroups allows partitioning
255a large system into nested, dynamically changeable, "soft-partitions".
256
257The attachment of each task, automatically inherited at fork by any
258children of that task, to a cgroup allows organizing the work load
259on a system into related sets of tasks. A task may be re-attached to
260any other cgroup, if allowed by the permissions on the necessary
261cgroup file system directories.
262
263When a task is moved from one cgroup to another, it gets a new
264css_set pointer - if there's an already existing css_set with the
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200265desired collection of cgroups then that group is reused, otherwise a new
Li Zefanb851ee72009-02-18 14:48:14 -0800266css_set is allocated. The appropriate existing css_set is located by
267looking into a hash table.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700268
Paul Menage817929e2007-10-18 23:39:36 -0700269To allow access from a cgroup to the css_sets (and hence tasks)
270that comprise it, a set of cg_cgroup_link objects form a lattice;
271each cg_cgroup_link is linked into a list of cg_cgroup_links for
Li Zefand19e0582008-02-23 15:24:08 -0800272a single cgroup on its cgrp_link_list field, and a list of
Paul Menage817929e2007-10-18 23:39:36 -0700273cg_cgroup_links for a single css_set on its cg_link_list.
274
275Thus the set of tasks in a cgroup can be listed by iterating over
276each css_set that references the cgroup, and sub-iterating over
277each css_set's task set.
278
Paul Menageddbcc7e2007-10-18 23:39:30 -0700279The use of a Linux virtual file system (vfs) to represent the
280cgroup hierarchy provides for a familiar permission and name space
281for cgroups, with a minimum of additional kernel code.
282
2831.4 What does notify_on_release do ?
284------------------------------------
285
Paul Menageddbcc7e2007-10-18 23:39:30 -0700286If the notify_on_release flag is enabled (1) in a cgroup, then
287whenever the last task in the cgroup leaves (exits or attaches to
288some other cgroup) and the last child cgroup of that cgroup
289is removed, then the kernel runs the command specified by the contents
290of the "release_agent" file in that hierarchy's root directory,
291supplying the pathname (relative to the mount point of the cgroup
292file system) of the abandoned cgroup. This enables automatic
293removal of abandoned cgroups. The default value of
294notify_on_release in the root cgroup at system boot is disabled
295(0). The default value of other cgroups at creation is the current
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200296value of their parents' notify_on_release settings. The default value of
Paul Menageddbcc7e2007-10-18 23:39:30 -0700297a cgroup hierarchy's release_agent path is empty.
298
Daniel Lezcano97978e62010-10-27 15:33:35 -07002991.5 What does clone_children do ?
300---------------------------------
301
Tejun Heo2260e7f2012-11-19 08:13:38 -0800302This flag only affects the cpuset controller. If the clone_children
303flag is enabled (1) in a cgroup, a new cpuset cgroup will copy its
304configuration from the parent during initialization.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700305
3061.6 How do I use cgroups ?
Paul Menageddbcc7e2007-10-18 23:39:30 -0700307--------------------------
308
309To start a new job that is to be contained within a cgroup, using
310the "cpuset" cgroup subsystem, the steps are something like:
311
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700312 1) mount -t tmpfs cgroup_root /sys/fs/cgroup
313 2) mkdir /sys/fs/cgroup/cpuset
314 3) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset
315 4) Create the new cgroup by doing mkdir's and write's (or echo's) in
316 the /sys/fs/cgroup virtual file system.
317 5) Start a task that will be the "founding father" of the new job.
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200318 6) Attach that task to the new cgroup by writing its PID to the
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700319 /sys/fs/cgroup/cpuset/tasks file for that cgroup.
320 7) fork, exec or clone the job tasks from this founding father task.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700321
322For example, the following sequence of commands will setup a cgroup
323named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
324and then start a subshell 'sh' in that cgroup:
325
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700326 mount -t tmpfs cgroup_root /sys/fs/cgroup
327 mkdir /sys/fs/cgroup/cpuset
328 mount -t cgroup cpuset -ocpuset /sys/fs/cgroup/cpuset
329 cd /sys/fs/cgroup/cpuset
Paul Menageddbcc7e2007-10-18 23:39:30 -0700330 mkdir Charlie
331 cd Charlie
Dhaval Giani0f146a72008-05-12 14:02:31 -0700332 /bin/echo 2-3 > cpuset.cpus
333 /bin/echo 1 > cpuset.mems
Paul Menageddbcc7e2007-10-18 23:39:30 -0700334 /bin/echo $$ > tasks
335 sh
336 # The subshell 'sh' is now running in cgroup Charlie
337 # The next line should display '/Charlie'
338 cat /proc/self/cgroup
339
3402. Usage Examples and Syntax
341============================
342
3432.1 Basic Usage
344---------------
345
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200346Creating, modifying, using cgroups can be done through the cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -0700347virtual filesystem.
348
Chris Samuelcaa790b2009-01-17 00:01:18 +1100349To mount a cgroup hierarchy with all available subsystems, type:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700350# mount -t cgroup xxx /sys/fs/cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -0700351
352The "xxx" is not interpreted by the cgroup code, but will appear in
353/proc/mounts so may be any useful identifying string that you like.
354
Eric B Munsonbb6405e2011-03-15 16:12:18 -0700355Note: Some subsystems do not work without some user input first. For instance,
356if cpusets are enabled the user will have to populate the cpus and mems files
357for each new cgroup created before that group can be used.
358
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700359As explained in section `1.2 Why are cgroups needed?' you should create
360different hierarchies of cgroups for each single resource or group of
361resources you want to control. Therefore, you should mount a tmpfs on
362/sys/fs/cgroup and create directories for each cgroup resource or resource
363group.
364
365# mount -t tmpfs cgroup_root /sys/fs/cgroup
366# mkdir /sys/fs/cgroup/rg1
367
Trevor Woerner595f4b62010-05-26 14:42:35 -0700368To mount a cgroup hierarchy with just the cpuset and memory
Paul Menageddbcc7e2007-10-18 23:39:30 -0700369subsystems, type:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700370# mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700371
Daniel Wagner9a8054a2012-07-10 10:49:18 +0200372While remounting cgroups is currently supported, it is not recommend
373to use it. Remounting allows changing bound subsystems and
374release_agent. Rebinding is hardly useful as it only works when the
375hierarchy is empty and release_agent itself should be replaced with
376conventional fsnotify. The support for remounting will be removed in
377the future.
Li Zefanb6719ec2009-04-02 16:57:28 -0700378
379To Specify a hierarchy's release_agent:
380# mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700381 xxx /sys/fs/cgroup/rg1
Li Zefanb6719ec2009-04-02 16:57:28 -0700382
383Note that specifying 'release_agent' more than once will return failure.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700384
Tejun Heo26d5bbe2013-04-12 10:29:04 -0700385Note that changing the set of subsystems is currently only supported
386when the hierarchy consists of a single (root) cgroup. Supporting
387the ability to arbitrarily bind/unbind subsystems from an existing
388cgroup hierarchy is intended to be implemented in the future.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700389
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700390Then under /sys/fs/cgroup/rg1 you can find a tree that corresponds to the
391tree of the cgroups in the system. For instance, /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700392is the cgroup that holds the whole system.
393
Li Zefanb6719ec2009-04-02 16:57:28 -0700394If you want to change the value of release_agent:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700395# echo "/sbin/new_release_agent" > /sys/fs/cgroup/rg1/release_agent
Li Zefanb6719ec2009-04-02 16:57:28 -0700396
397It can also be changed via remount.
398
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700399If you want to create a new cgroup under /sys/fs/cgroup/rg1:
400# cd /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700401# mkdir my_cgroup
402
403Now you want to do something with this cgroup.
404# cd my_cgroup
405
406In this directory you can find several files:
407# ls
Paul Menage7823da32009-10-07 16:32:26 -0700408cgroup.procs notify_on_release tasks
Li Zefand19e0582008-02-23 15:24:08 -0800409(plus whatever files added by the attached subsystems)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700410
411Now attach your shell to this cgroup:
412# /bin/echo $$ > tasks
413
414You can also create cgroups inside your cgroup by using mkdir in this
415directory.
416# mkdir my_sub_cs
417
418To remove a cgroup, just use rmdir:
419# rmdir my_sub_cs
420
421This will fail if the cgroup is in use (has cgroups inside, or
422has processes attached, or is held alive by other subsystem-specific
423reference).
424
4252.2 Attaching processes
426-----------------------
427
428# /bin/echo PID > tasks
429
430Note that it is PID, not PIDs. You can only attach ONE task at a time.
431If you have several tasks to attach, you have to do it one after another:
432
433# /bin/echo PID1 > tasks
434# /bin/echo PID2 > tasks
435 ...
436# /bin/echo PIDn > tasks
437
Li Zefanbef67c52008-07-04 09:59:55 -0700438You can attach the current shell task by echoing 0:
439
440# echo 0 > tasks
441
Ben Blum74a11662011-05-26 16:25:20 -0700442You can use the cgroup.procs file instead of the tasks file to move all
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200443threads in a threadgroup at once. Echoing the PID of any task in a
Ben Blum74a11662011-05-26 16:25:20 -0700444threadgroup to cgroup.procs causes all tasks in that threadgroup to be
Rami Rosen1ae65ae2013-04-03 20:32:04 +0300445attached to the cgroup. Writing 0 to cgroup.procs moves all tasks
Ben Blum74a11662011-05-26 16:25:20 -0700446in the writing task's threadgroup.
447
Eric B Munsonbb6405e2011-03-15 16:12:18 -0700448Note: Since every task is always a member of exactly one cgroup in each
449mounted hierarchy, to remove a task from its current cgroup you must
450move it into a new cgroup (possibly the root cgroup) by writing to the
451new cgroup's tasks file.
452
Li Zefan5fe69d72011-11-04 11:22:05 -0700453Note: Due to some restrictions enforced by some cgroup subsystems, moving
454a process to another cgroup can fail.
Eric B Munsonbb6405e2011-03-15 16:12:18 -0700455
Paul Menagec6d57f32009-09-23 15:56:19 -07004562.3 Mounting hierarchies by name
457--------------------------------
458
459Passing the name=<x> option when mounting a cgroups hierarchy
460associates the given name with the hierarchy. This can be used when
461mounting a pre-existing hierarchy, in order to refer to it by name
462rather than by its set of active subsystems. Each hierarchy is either
463nameless, or has a unique name.
464
465The name should match [\w.-]+
466
467When passing a name=<x> option for a new hierarchy, you need to
468specify subsystems manually; the legacy behaviour of mounting all
469subsystems when none are explicitly specified is not supported when
470you give a subsystem a name.
471
472The name of the subsystem appears as part of the hierarchy description
473in /proc/mounts and /proc/<pid>/cgroups.
474
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08004752.4 Notification API
476--------------------
477
478There is mechanism which allows to get notifications about changing
479status of a cgroup.
480
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200481To register a new notification handler you need to:
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800482 - create a file descriptor for event notification using eventfd(2);
483 - open a control file to be monitored (e.g. memory.usage_in_bytes);
484 - write "<event_fd> <control_fd> <args>" to cgroup.event_control.
485 Interpretation of args is defined by control file implementation;
486
487eventfd will be woken up by control file implementation or when the
488cgroup is removed.
489
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200490To unregister a notification handler just close eventfd.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800491
492NOTE: Support of notifications should be implemented for the control
493file. See documentation for the subsystem.
Paul Menagec6d57f32009-09-23 15:56:19 -0700494
Paul Menageddbcc7e2007-10-18 23:39:30 -07004953. Kernel API
496=============
497
4983.1 Overview
499------------
500
501Each kernel subsystem that wants to hook into the generic cgroup
502system needs to create a cgroup_subsys object. This contains
503various methods, which are callbacks from the cgroup system, along
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200504with a subsystem ID which will be assigned by the cgroup system.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700505
506Other fields in the cgroup_subsys object include:
507
508- subsys_id: a unique array index for the subsystem, indicating which
Li Zefand19e0582008-02-23 15:24:08 -0800509 entry in cgroup->subsys[] this subsystem should be managing.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700510
Li Zefand19e0582008-02-23 15:24:08 -0800511- name: should be initialized to a unique subsystem name. Should be
512 no longer than MAX_CGROUP_TYPE_NAMELEN.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700513
Li Zefand19e0582008-02-23 15:24:08 -0800514- early_init: indicate if the subsystem needs early initialization
515 at system boot.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700516
517Each cgroup object created by the system has an array of pointers,
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200518indexed by subsystem ID; this pointer is entirely managed by the
Paul Menageddbcc7e2007-10-18 23:39:30 -0700519subsystem; the generic cgroup code will never touch this pointer.
520
5213.2 Synchronization
522-------------------
523
524There is a global mutex, cgroup_mutex, used by the cgroup
525system. This should be taken by anything that wants to modify a
526cgroup. It may also be taken to prevent cgroups from being
527modified, but more specific locks may be more appropriate in that
528situation.
529
530See kernel/cgroup.c for more details.
531
532Subsystems can take/release the cgroup_mutex via the functions
Paul Menageddbcc7e2007-10-18 23:39:30 -0700533cgroup_lock()/cgroup_unlock().
534
535Accessing a task's cgroup pointer may be done in the following ways:
536- while holding cgroup_mutex
537- while holding the task's alloc_lock (via task_lock())
538- inside an rcu_read_lock() section via rcu_dereference()
539
5403.3 Subsystem API
Li Zefand19e0582008-02-23 15:24:08 -0800541-----------------
Paul Menageddbcc7e2007-10-18 23:39:30 -0700542
543Each subsystem should:
544
545- add an entry in linux/cgroup_subsys.h
546- define a cgroup_subsys object called <name>_subsys
547
Ben Blume6a11052010-03-10 15:22:09 -0800548If a subsystem can be compiled as a module, it should also have in its
Ben Blumcf5d5942010-03-10 15:22:09 -0800549module initcall a call to cgroup_load_subsys(), and in its exitcall a
550call to cgroup_unload_subsys(). It should also set its_subsys.module =
551THIS_MODULE in its .c file.
Ben Blume6a11052010-03-10 15:22:09 -0800552
Paul Menageddbcc7e2007-10-18 23:39:30 -0700553Each subsystem may export the following methods. The only mandatory
Tejun Heo92fb9742012-11-19 08:13:38 -0800554methods are css_alloc/free. Any others that are null are presumed to
Paul Menageddbcc7e2007-10-18 23:39:30 -0700555be successful no-ops.
556
Tejun Heo92fb9742012-11-19 08:13:38 -0800557struct cgroup_subsys_state *css_alloc(struct cgroup *cgrp)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800558(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700559
Tejun Heo92fb9742012-11-19 08:13:38 -0800560Called to allocate a subsystem state object for a cgroup. The
Paul Menageddbcc7e2007-10-18 23:39:30 -0700561subsystem should allocate its subsystem state object for the passed
562cgroup, returning a pointer to the new object on success or a
Tejun Heo92fb9742012-11-19 08:13:38 -0800563ERR_PTR() value. On success, the subsystem pointer should point to
Paul Menageddbcc7e2007-10-18 23:39:30 -0700564a structure of type cgroup_subsys_state (typically embedded in a
565larger subsystem-specific object), which will be initialized by the
566cgroup system. Note that this will be called at initialization to
567create the root subsystem state for this subsystem; this case can be
568identified by the passed cgroup object having a NULL parent (since
569it's the root of the hierarchy) and may be an appropriate place for
570initialization code.
571
Tejun Heo92fb9742012-11-19 08:13:38 -0800572int css_online(struct cgroup *cgrp)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800573(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700574
Tejun Heo92fb9742012-11-19 08:13:38 -0800575Called after @cgrp successfully completed all allocations and made
576visible to cgroup_for_each_child/descendant_*() iterators. The
577subsystem may choose to fail creation by returning -errno. This
578callback can be used to implement reliable state sharing and
579propagation along the hierarchy. See the comment on
580cgroup_for_each_descendant_pre() for details.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700581
Tejun Heo92fb9742012-11-19 08:13:38 -0800582void css_offline(struct cgroup *cgrp);
Li Zefand7eeac12013-03-12 15:35:59 -0700583(cgroup_mutex held by caller)
Li Zefand19e0582008-02-23 15:24:08 -0800584
Tejun Heo92fb9742012-11-19 08:13:38 -0800585This is the counterpart of css_online() and called iff css_online()
586has succeeded on @cgrp. This signifies the beginning of the end of
587@cgrp. @cgrp is being removed and the subsystem should start dropping
588all references it's holding on @cgrp. When all references are dropped,
589cgroup removal will proceed to the next step - css_free(). After this
590callback, @cgrp should be considered dead to the subsystem.
591
592void css_free(struct cgroup *cgrp)
593(cgroup_mutex held by caller)
594
595The cgroup system is about to free @cgrp; the subsystem should free
596its subsystem state object. By the time this method is called, @cgrp
597is completely unused; @cgrp->parent is still valid. (Note - can also
598be called for a newly-created cgroup if an error occurs after this
599subsystem's create() method has been called for the new cgroup).
Li Zefand19e0582008-02-23 15:24:08 -0800600
Li Zefan761b3ef2012-01-31 13:47:36 +0800601int can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800602(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700603
Tejun Heo2f7ee562011-12-12 18:12:21 -0800604Called prior to moving one or more tasks into a cgroup; if the
605subsystem returns an error, this will abort the attach operation.
606@tset contains the tasks to be attached and is guaranteed to have at
607least one task in it.
608
609If there are multiple tasks in the taskset, then:
610 - it's guaranteed that all are from the same thread group
611 - @tset contains all tasks from the thread group whether or not
612 they're switching cgroups
613 - the first task is the leader
614
615Each @tset entry also contains the task's old cgroup and tasks which
616aren't switching cgroup can be skipped easily using the
617cgroup_taskset_for_each() iterator. Note that this isn't called on a
618fork. If this method returns 0 (success) then this should remain valid
619while the caller holds cgroup_mutex and it is ensured that either
Ben Blumf780bdb2011-05-26 16:25:19 -0700620attach() or cancel_attach() will be called in future.
621
Li Zefan761b3ef2012-01-31 13:47:36 +0800622void cancel_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800623(cgroup_mutex held by caller)
624
625Called when a task attach operation has failed after can_attach() has succeeded.
626A subsystem whose can_attach() has some side-effects should provide this
Thomas Weber88393162010-03-16 11:47:56 +0100627function, so that the subsystem can implement a rollback. If not, not necessary.
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800628This will be called only about subsystems whose can_attach() operation have
Tejun Heo2f7ee562011-12-12 18:12:21 -0800629succeeded. The parameters are identical to can_attach().
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800630
Li Zefan761b3ef2012-01-31 13:47:36 +0800631void attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Li Zefan18e7f1f2009-01-07 18:07:32 -0800632(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700633
634Called after the task has been attached to the cgroup, to allow any
635post-attachment activity that requires memory allocations or blocking.
Tejun Heo2f7ee562011-12-12 18:12:21 -0800636The parameters are identical to can_attach().
Ben Blumf780bdb2011-05-26 16:25:19 -0700637
Li Zefan761b3ef2012-01-31 13:47:36 +0800638void fork(struct task_struct *task)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700639
Li Zefane8d55fd2008-04-29 01:00:13 -0700640Called when a task is forked into a cgroup.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700641
Li Zefan761b3ef2012-01-31 13:47:36 +0800642void exit(struct task_struct *task)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700643
Li Zefand19e0582008-02-23 15:24:08 -0800644Called during task exit.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700645
Tejun Heo26d5bbe2013-04-12 10:29:04 -0700646void bind(struct cgroup *root)
647(cgroup_mutex held by caller)
648
649Called when a cgroup subsystem is rebound to a different hierarchy
650and root cgroup. Currently this will only involve movement between
651the default hierarchy (which never has sub-cgroups) and a hierarchy
652that is being created/destroyed (and hence has no sub-cgroups).
Paul Menageddbcc7e2007-10-18 23:39:30 -0700653
Aristeu Rozanski19ec2562012-09-11 16:28:10 -04006544. Extended attribute usage
655===========================
656
657cgroup filesystem supports certain types of extended attributes in its
658directories and files. The current supported types are:
659 - Trusted (XATTR_TRUSTED)
660 - Security (XATTR_SECURITY)
661
662Both require CAP_SYS_ADMIN capability to set.
663
664Like in tmpfs, the extended attributes in cgroup filesystem are stored
665using kernel memory and it's advised to keep the usage at minimum. This
666is the reason why user defined extended attributes are not supported, since
667any user can do it and there's no limit in the value size.
668
669The current known users for this feature are SELinux to limit cgroup usage
670in containers and systemd for assorted meta data like main PID in a cgroup
671(systemd creates a cgroup per service).
672
6735. Questions
Paul Menageddbcc7e2007-10-18 23:39:30 -0700674============
675
676Q: what's up with this '/bin/echo' ?
677A: bash's builtin 'echo' command does not check calls to write() against
678 errors. If you use it in the cgroup file system, you won't be
679 able to tell whether a command succeeded or failed.
680
681Q: When I attach processes, only the first of the line gets really attached !
682A: We can only return one error code per call to write(). So you should also
Michael Kerrisk83b061f2012-09-11 13:20:20 +0200683 put only ONE PID.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700684