blob: cd67e90003c0e547952de41086f01d59ecb9a493 [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
324. Questions
33
341. Control Groups
Li Zefand19e0582008-02-23 15:24:08 -080035=================
Paul Menageddbcc7e2007-10-18 23:39:30 -070036
371.1 What are cgroups ?
38----------------------
39
40Control Groups provide a mechanism for aggregating/partitioning sets of
41tasks, and all their future children, into hierarchical groups with
42specialized behaviour.
43
44Definitions:
45
46A *cgroup* associates a set of tasks with a set of parameters for one
47or more subsystems.
48
49A *subsystem* is a module that makes use of the task grouping
50facilities provided by cgroups to treat groups of tasks in
51particular ways. A subsystem is typically a "resource controller" that
52schedules a resource or applies per-cgroup limits, but it may be
53anything that wants to act on a group of processes, e.g. a
54virtualization subsystem.
55
56A *hierarchy* is a set of cgroups arranged in a tree, such that
57every task in the system is in exactly one of the cgroups in the
58hierarchy, and a set of subsystems; each subsystem has system-specific
59state attached to each cgroup in the hierarchy. Each hierarchy has
60an instance of the cgroup virtual filesystem associated with it.
61
Chris Samuelcaa790b2009-01-17 00:01:18 +110062At any one time there may be multiple active hierarchies of task
Paul Menageddbcc7e2007-10-18 23:39:30 -070063cgroups. Each hierarchy is a partition of all tasks in the system.
64
65User level code may create and destroy cgroups by name in an
66instance of the cgroup virtual file system, specify and query to
67which cgroup a task is assigned, and list the task pids assigned to
68a cgroup. Those creations and assignments only affect the hierarchy
69associated with that instance of the cgroup file system.
70
71On their own, the only use for cgroups is for simple job
72tracking. The intention is that other subsystems hook into the generic
73cgroup support to provide new attributes for cgroups, such as
74accounting/limiting the resources which processes in a cgroup can
Li Zefan45ce80f2009-01-15 13:50:59 -080075access. For example, cpusets (see Documentation/cgroups/cpusets.txt) allows
Paul Menageddbcc7e2007-10-18 23:39:30 -070076you to associate a set of CPUs and a set of memory nodes with the
77tasks in each cgroup.
78
791.2 Why are cgroups needed ?
80----------------------------
81
82There are multiple efforts to provide process aggregations in the
83Linux kernel, mainly for resource tracking purposes. Such efforts
84include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server
85namespaces. These all require the basic notion of a
86grouping/partitioning of processes, with newly forked processes ending
87in the same group (cgroup) as their parent process.
88
89The kernel cgroup patch provides the minimum essential kernel
90mechanisms required to efficiently implement such groups. It has
91minimal impact on the system fast paths, and provides hooks for
92specific subsystems such as cpusets to provide additional behaviour as
93desired.
94
95Multiple hierarchy support is provided to allow for situations where
96the division of tasks into cgroups is distinctly different for
97different subsystems - having parallel hierarchies allows each
98hierarchy to be a natural division of tasks, without having to handle
99complex combinations of tasks that would be present if several
100unrelated subsystems needed to be forced into the same tree of
101cgroups.
102
103At one extreme, each resource controller or subsystem could be in a
104separate hierarchy; at the other extreme, all subsystems
105would be attached to the same hierarchy.
106
107As an example of a scenario (originally proposed by vatsa@in.ibm.com)
108that can benefit from multiple hierarchies, consider a large
109university server with various users - students, professors, system
110tasks etc. The resource planning for this server could be along the
111following lines:
112
Geunsik Lim6ad85232011-04-04 15:10:45 -0700113 CPU : "Top cpuset"
Paul Menageddbcc7e2007-10-18 23:39:30 -0700114 / \
115 CPUSet1 CPUSet2
Geunsik Lim6ad85232011-04-04 15:10:45 -0700116 | |
117 (Professors) (Students)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700118
119 In addition (system tasks) are attached to topcpuset (so
120 that they can run anywhere) with a limit of 20%
121
Geunsik Lim6ad85232011-04-04 15:10:45 -0700122 Memory : Professors (50%), Students (30%), system (20%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700123
Geunsik Lim6ad85232011-04-04 15:10:45 -0700124 Disk : Professors (50%), Students (30%), system (20%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700125
126 Network : WWW browsing (20%), Network File System (60%), others (20%)
127 / \
Geunsik Lim6ad85232011-04-04 15:10:45 -0700128 Professors (15%) students (5%)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700129
Chris Samuelcaa790b2009-01-17 00:01:18 +1100130Browsers like Firefox/Lynx go into the WWW network class, while (k)nfsd go
Paul Menageddbcc7e2007-10-18 23:39:30 -0700131into NFS network class.
132
Chris Samuelcaa790b2009-01-17 00:01:18 +1100133At the same time Firefox/Lynx will share an appropriate CPU/Memory class
Paul Menageddbcc7e2007-10-18 23:39:30 -0700134depending on who launched it (prof/student).
135
136With the ability to classify tasks differently for different resources
137(by putting those resource subsystems in different hierarchies) then
138the admin can easily set up a script which receives exec notifications
139and depending on who is launching the browser he can
140
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700141 # echo browser_pid > /sys/fs/cgroup/<restype>/<userclass>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700142
143With only a single hierarchy, he now would potentially have to create
144a separate cgroup for every browser launched and associate it with
Jörg Sommer67de0162011-06-15 13:00:47 -0700145appropriate network and other resource class. This may lead to
Paul Menageddbcc7e2007-10-18 23:39:30 -0700146proliferation of such cgroups.
147
148Also lets say that the administrator would like to give enhanced network
149access temporarily to a student's browser (since it is night and the user
Li Zefand19e0582008-02-23 15:24:08 -0800150wants to do online gaming :)) OR give one of the students simulation
Paul Menageddbcc7e2007-10-18 23:39:30 -0700151apps enhanced CPU power,
152
Li Zefand19e0582008-02-23 15:24:08 -0800153With ability to write pids directly to resource classes, it's just a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700154matter of :
155
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700156 # echo pid > /sys/fs/cgroup/network/<new_class>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700157 (after some time)
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700158 # echo pid > /sys/fs/cgroup/network/<orig_class>/tasks
Paul Menageddbcc7e2007-10-18 23:39:30 -0700159
160Without this ability, he would have to split the cgroup into
161multiple separate ones and then associate the new cgroups with the
162new resource classes.
163
164
165
1661.3 How are cgroups implemented ?
167---------------------------------
168
169Control Groups extends the kernel as follows:
170
171 - Each task in the system has a reference-counted pointer to a
172 css_set.
173
174 - A css_set contains a set of reference-counted pointers to
175 cgroup_subsys_state objects, one for each cgroup subsystem
176 registered in the system. There is no direct link from a task to
177 the cgroup of which it's a member in each hierarchy, but this
178 can be determined by following pointers through the
179 cgroup_subsys_state objects. This is because accessing the
180 subsystem state is something that's expected to happen frequently
181 and in performance-critical code, whereas operations that require a
182 task's actual cgroup assignments (in particular, moving between
Paul Menage817929e2007-10-18 23:39:36 -0700183 cgroups) are less common. A linked list runs through the cg_list
184 field of each task_struct using the css_set, anchored at
185 css_set->tasks.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700186
187 - A cgroup hierarchy filesystem can be mounted for browsing and
188 manipulation from user space.
189
190 - You can list all the tasks (by pid) attached to any cgroup.
191
192The implementation of cgroups requires a few, simple hooks
193into the rest of the kernel, none in performance critical paths:
194
195 - in init/main.c, to initialize the root cgroups and initial
196 css_set at system boot.
197
198 - in fork and exit, to attach and detach a task from its css_set.
199
200In addition a new file system, of type "cgroup" may be mounted, to
201enable browsing and modifying the cgroups presently known to the
202kernel. When mounting a cgroup hierarchy, you may specify a
203comma-separated list of subsystems to mount as the filesystem mount
204options. By default, mounting the cgroup filesystem attempts to
205mount a hierarchy containing all registered subsystems.
206
207If an active hierarchy with exactly the same set of subsystems already
208exists, it will be reused for the new mount. If no existing hierarchy
209matches, and any of the requested subsystems are in use in an existing
210hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy
211is activated, associated with the requested subsystems.
212
213It's not currently possible to bind a new subsystem to an active
214cgroup hierarchy, or to unbind a subsystem from an active cgroup
215hierarchy. This may be possible in future, but is fraught with nasty
216error-recovery issues.
217
218When a cgroup filesystem is unmounted, if there are any
219child cgroups created below the top-level cgroup, that hierarchy
220will remain active even though unmounted; if there are no
221child cgroups then the hierarchy will be deactivated.
222
223No new system calls are added for cgroups - all support for
224querying and modifying cgroups is via this cgroup file system.
225
226Each task under /proc has an added file named 'cgroup' displaying,
227for each active hierarchy, the subsystem names and the cgroup name
228as the path relative to the root of the cgroup file system.
229
230Each cgroup is represented by a directory in the cgroup file system
231containing the following files describing that cgroup:
232
Paul Menage7823da32009-10-07 16:32:26 -0700233 - tasks: list of tasks (by pid) attached to that cgroup. This list
234 is not guaranteed to be sorted. Writing a thread id into this file
235 moves the thread into this cgroup.
236 - cgroup.procs: list of tgids in the cgroup. This list is not
237 guaranteed to be sorted or free of duplicate tgids, and userspace
238 should sort/uniquify the list if this property is required.
Ben Blum74a11662011-05-26 16:25:20 -0700239 Writing a thread group id into this file moves all threads in that
240 group into this cgroup.
Li Zefand19e0582008-02-23 15:24:08 -0800241 - notify_on_release flag: run the release agent on exit?
242 - release_agent: the path to use for release notifications (this file
243 exists in the top cgroup only)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700244
245Other subsystems such as cpusets may add additional files in each
Li Zefand19e0582008-02-23 15:24:08 -0800246cgroup dir.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700247
248New cgroups are created using the mkdir system call or shell
249command. The properties of a cgroup, such as its flags, are
250modified by writing to the appropriate file in that cgroups
251directory, as listed above.
252
253The named hierarchical structure of nested cgroups allows partitioning
254a large system into nested, dynamically changeable, "soft-partitions".
255
256The attachment of each task, automatically inherited at fork by any
257children of that task, to a cgroup allows organizing the work load
258on a system into related sets of tasks. A task may be re-attached to
259any other cgroup, if allowed by the permissions on the necessary
260cgroup file system directories.
261
262When a task is moved from one cgroup to another, it gets a new
263css_set pointer - if there's an already existing css_set with the
264desired collection of cgroups then that group is reused, else a new
Li Zefanb851ee72009-02-18 14:48:14 -0800265css_set is allocated. The appropriate existing css_set is located by
266looking into a hash table.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700267
Paul Menage817929e2007-10-18 23:39:36 -0700268To allow access from a cgroup to the css_sets (and hence tasks)
269that comprise it, a set of cg_cgroup_link objects form a lattice;
270each cg_cgroup_link is linked into a list of cg_cgroup_links for
Li Zefand19e0582008-02-23 15:24:08 -0800271a single cgroup on its cgrp_link_list field, and a list of
Paul Menage817929e2007-10-18 23:39:36 -0700272cg_cgroup_links for a single css_set on its cg_link_list.
273
274Thus the set of tasks in a cgroup can be listed by iterating over
275each css_set that references the cgroup, and sub-iterating over
276each css_set's task set.
277
Paul Menageddbcc7e2007-10-18 23:39:30 -0700278The use of a Linux virtual file system (vfs) to represent the
279cgroup hierarchy provides for a familiar permission and name space
280for cgroups, with a minimum of additional kernel code.
281
2821.4 What does notify_on_release do ?
283------------------------------------
284
Paul Menageddbcc7e2007-10-18 23:39:30 -0700285If the notify_on_release flag is enabled (1) in a cgroup, then
286whenever the last task in the cgroup leaves (exits or attaches to
287some other cgroup) and the last child cgroup of that cgroup
288is removed, then the kernel runs the command specified by the contents
289of the "release_agent" file in that hierarchy's root directory,
290supplying the pathname (relative to the mount point of the cgroup
291file system) of the abandoned cgroup. This enables automatic
292removal of abandoned cgroups. The default value of
293notify_on_release in the root cgroup at system boot is disabled
294(0). The default value of other cgroups at creation is the current
295value of their parents notify_on_release setting. The default value of
296a cgroup hierarchy's release_agent path is empty.
297
Daniel Lezcano97978e62010-10-27 15:33:35 -07002981.5 What does clone_children do ?
299---------------------------------
300
301If the clone_children flag is enabled (1) in a cgroup, then all
302cgroups created beneath will call the post_clone callbacks for each
303subsystem of the newly created cgroup. Usually when this callback is
304implemented for a subsystem, it copies the values of the parent
305subsystem, this is the case for the cpuset.
306
3071.6 How do I use cgroups ?
Paul Menageddbcc7e2007-10-18 23:39:30 -0700308--------------------------
309
310To start a new job that is to be contained within a cgroup, using
311the "cpuset" cgroup subsystem, the steps are something like:
312
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700313 1) mount -t tmpfs cgroup_root /sys/fs/cgroup
314 2) mkdir /sys/fs/cgroup/cpuset
315 3) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset
316 4) Create the new cgroup by doing mkdir's and write's (or echo's) in
317 the /sys/fs/cgroup virtual file system.
318 5) Start a task that will be the "founding father" of the new job.
319 6) Attach that task to the new cgroup by writing its pid to the
320 /sys/fs/cgroup/cpuset/tasks file for that cgroup.
321 7) fork, exec or clone the job tasks from this founding father task.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700322
323For example, the following sequence of commands will setup a cgroup
324named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
325and then start a subshell 'sh' in that cgroup:
326
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700327 mount -t tmpfs cgroup_root /sys/fs/cgroup
328 mkdir /sys/fs/cgroup/cpuset
329 mount -t cgroup cpuset -ocpuset /sys/fs/cgroup/cpuset
330 cd /sys/fs/cgroup/cpuset
Paul Menageddbcc7e2007-10-18 23:39:30 -0700331 mkdir Charlie
332 cd Charlie
Dhaval Giani0f146a72008-05-12 14:02:31 -0700333 /bin/echo 2-3 > cpuset.cpus
334 /bin/echo 1 > cpuset.mems
Paul Menageddbcc7e2007-10-18 23:39:30 -0700335 /bin/echo $$ > tasks
336 sh
337 # The subshell 'sh' is now running in cgroup Charlie
338 # The next line should display '/Charlie'
339 cat /proc/self/cgroup
340
3412. Usage Examples and Syntax
342============================
343
3442.1 Basic Usage
345---------------
346
347Creating, modifying, using the cgroups can be done through the cgroup
348virtual filesystem.
349
Chris Samuelcaa790b2009-01-17 00:01:18 +1100350To mount a cgroup hierarchy with all available subsystems, type:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700351# mount -t cgroup xxx /sys/fs/cgroup
Paul Menageddbcc7e2007-10-18 23:39:30 -0700352
353The "xxx" is not interpreted by the cgroup code, but will appear in
354/proc/mounts so may be any useful identifying string that you like.
355
Eric B Munsonbb6405e2011-03-15 16:12:18 -0700356Note: Some subsystems do not work without some user input first. For instance,
357if cpusets are enabled the user will have to populate the cpus and mems files
358for each new cgroup created before that group can be used.
359
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700360As explained in section `1.2 Why are cgroups needed?' you should create
361different hierarchies of cgroups for each single resource or group of
362resources you want to control. Therefore, you should mount a tmpfs on
363/sys/fs/cgroup and create directories for each cgroup resource or resource
364group.
365
366# mount -t tmpfs cgroup_root /sys/fs/cgroup
367# mkdir /sys/fs/cgroup/rg1
368
Trevor Woerner595f4b62010-05-26 14:42:35 -0700369To mount a cgroup hierarchy with just the cpuset and memory
Paul Menageddbcc7e2007-10-18 23:39:30 -0700370subsystems, type:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700371# mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700372
373To change the set of subsystems bound to a mounted hierarchy, just
374remount with different options:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700375# mount -o remount,cpuset,blkio hier1 /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700376
Trevor Woerner1bdcd782011-01-12 17:00:29 -0800377Now memory is removed from the hierarchy and blkio is added.
Li Zefanb6719ec2009-04-02 16:57:28 -0700378
Trevor Woerner1bdcd782011-01-12 17:00:29 -0800379Note this will add blkio to the hierarchy but won't remove memory or
Li Zefanb6719ec2009-04-02 16:57:28 -0700380cpuset, because the new options are appended to the old ones:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700381# mount -o remount,blkio /sys/fs/cgroup/rg1
Li Zefanb6719ec2009-04-02 16:57:28 -0700382
383To Specify a hierarchy's release_agent:
384# mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700385 xxx /sys/fs/cgroup/rg1
Li Zefanb6719ec2009-04-02 16:57:28 -0700386
387Note that specifying 'release_agent' more than once will return failure.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700388
389Note that changing the set of subsystems is currently only supported
390when the hierarchy consists of a single (root) cgroup. Supporting
391the ability to arbitrarily bind/unbind subsystems from an existing
392cgroup hierarchy is intended to be implemented in the future.
393
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700394Then under /sys/fs/cgroup/rg1 you can find a tree that corresponds to the
395tree of the cgroups in the system. For instance, /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700396is the cgroup that holds the whole system.
397
Li Zefanb6719ec2009-04-02 16:57:28 -0700398If you want to change the value of release_agent:
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700399# echo "/sbin/new_release_agent" > /sys/fs/cgroup/rg1/release_agent
Li Zefanb6719ec2009-04-02 16:57:28 -0700400
401It can also be changed via remount.
402
Jörg Sommerf6e07d32011-06-15 12:59:45 -0700403If you want to create a new cgroup under /sys/fs/cgroup/rg1:
404# cd /sys/fs/cgroup/rg1
Paul Menageddbcc7e2007-10-18 23:39:30 -0700405# mkdir my_cgroup
406
407Now you want to do something with this cgroup.
408# cd my_cgroup
409
410In this directory you can find several files:
411# ls
Paul Menage7823da32009-10-07 16:32:26 -0700412cgroup.procs notify_on_release tasks
Li Zefand19e0582008-02-23 15:24:08 -0800413(plus whatever files added by the attached subsystems)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700414
415Now attach your shell to this cgroup:
416# /bin/echo $$ > tasks
417
418You can also create cgroups inside your cgroup by using mkdir in this
419directory.
420# mkdir my_sub_cs
421
422To remove a cgroup, just use rmdir:
423# rmdir my_sub_cs
424
425This will fail if the cgroup is in use (has cgroups inside, or
426has processes attached, or is held alive by other subsystem-specific
427reference).
428
4292.2 Attaching processes
430-----------------------
431
432# /bin/echo PID > tasks
433
434Note that it is PID, not PIDs. You can only attach ONE task at a time.
435If you have several tasks to attach, you have to do it one after another:
436
437# /bin/echo PID1 > tasks
438# /bin/echo PID2 > tasks
439 ...
440# /bin/echo PIDn > tasks
441
Li Zefanbef67c52008-07-04 09:59:55 -0700442You can attach the current shell task by echoing 0:
443
444# echo 0 > tasks
445
Ben Blum74a11662011-05-26 16:25:20 -0700446You can use the cgroup.procs file instead of the tasks file to move all
447threads in a threadgroup at once. Echoing the pid of any task in a
448threadgroup to cgroup.procs causes all tasks in that threadgroup to be
449be attached to the cgroup. Writing 0 to cgroup.procs moves all tasks
450in the writing task's threadgroup.
451
Eric B Munsonbb6405e2011-03-15 16:12:18 -0700452Note: Since every task is always a member of exactly one cgroup in each
453mounted hierarchy, to remove a task from its current cgroup you must
454move it into a new cgroup (possibly the root cgroup) by writing to the
455new cgroup's tasks file.
456
457Note: If the ns cgroup is active, moving a process to another cgroup can
458fail.
459
Paul Menagec6d57f32009-09-23 15:56:19 -07004602.3 Mounting hierarchies by name
461--------------------------------
462
463Passing the name=<x> option when mounting a cgroups hierarchy
464associates the given name with the hierarchy. This can be used when
465mounting a pre-existing hierarchy, in order to refer to it by name
466rather than by its set of active subsystems. Each hierarchy is either
467nameless, or has a unique name.
468
469The name should match [\w.-]+
470
471When passing a name=<x> option for a new hierarchy, you need to
472specify subsystems manually; the legacy behaviour of mounting all
473subsystems when none are explicitly specified is not supported when
474you give a subsystem a name.
475
476The name of the subsystem appears as part of the hierarchy description
477in /proc/mounts and /proc/<pid>/cgroups.
478
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08004792.4 Notification API
480--------------------
481
482There is mechanism which allows to get notifications about changing
483status of a cgroup.
484
485To register new notification handler you need:
486 - create a file descriptor for event notification using eventfd(2);
487 - open a control file to be monitored (e.g. memory.usage_in_bytes);
488 - write "<event_fd> <control_fd> <args>" to cgroup.event_control.
489 Interpretation of args is defined by control file implementation;
490
491eventfd will be woken up by control file implementation or when the
492cgroup is removed.
493
494To unregister notification handler just close eventfd.
495
496NOTE: Support of notifications should be implemented for the control
497file. See documentation for the subsystem.
Paul Menagec6d57f32009-09-23 15:56:19 -0700498
Paul Menageddbcc7e2007-10-18 23:39:30 -07004993. Kernel API
500=============
501
5023.1 Overview
503------------
504
505Each kernel subsystem that wants to hook into the generic cgroup
506system needs to create a cgroup_subsys object. This contains
507various methods, which are callbacks from the cgroup system, along
508with a subsystem id which will be assigned by the cgroup system.
509
510Other fields in the cgroup_subsys object include:
511
512- subsys_id: a unique array index for the subsystem, indicating which
Li Zefand19e0582008-02-23 15:24:08 -0800513 entry in cgroup->subsys[] this subsystem should be managing.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700514
Li Zefand19e0582008-02-23 15:24:08 -0800515- name: should be initialized to a unique subsystem name. Should be
516 no longer than MAX_CGROUP_TYPE_NAMELEN.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700517
Li Zefand19e0582008-02-23 15:24:08 -0800518- early_init: indicate if the subsystem needs early initialization
519 at system boot.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700520
521Each cgroup object created by the system has an array of pointers,
522indexed by subsystem id; this pointer is entirely managed by the
523subsystem; the generic cgroup code will never touch this pointer.
524
5253.2 Synchronization
526-------------------
527
528There is a global mutex, cgroup_mutex, used by the cgroup
529system. This should be taken by anything that wants to modify a
530cgroup. It may also be taken to prevent cgroups from being
531modified, but more specific locks may be more appropriate in that
532situation.
533
534See kernel/cgroup.c for more details.
535
536Subsystems can take/release the cgroup_mutex via the functions
Paul Menageddbcc7e2007-10-18 23:39:30 -0700537cgroup_lock()/cgroup_unlock().
538
539Accessing a task's cgroup pointer may be done in the following ways:
540- while holding cgroup_mutex
541- while holding the task's alloc_lock (via task_lock())
542- inside an rcu_read_lock() section via rcu_dereference()
543
5443.3 Subsystem API
Li Zefand19e0582008-02-23 15:24:08 -0800545-----------------
Paul Menageddbcc7e2007-10-18 23:39:30 -0700546
547Each subsystem should:
548
549- add an entry in linux/cgroup_subsys.h
550- define a cgroup_subsys object called <name>_subsys
551
Ben Blume6a11052010-03-10 15:22:09 -0800552If a subsystem can be compiled as a module, it should also have in its
Ben Blumcf5d5942010-03-10 15:22:09 -0800553module initcall a call to cgroup_load_subsys(), and in its exitcall a
554call to cgroup_unload_subsys(). It should also set its_subsys.module =
555THIS_MODULE in its .c file.
Ben Blume6a11052010-03-10 15:22:09 -0800556
Paul Menageddbcc7e2007-10-18 23:39:30 -0700557Each subsystem may export the following methods. The only mandatory
558methods are create/destroy. Any others that are null are presumed to
559be successful no-ops.
560
Li Zefand19e0582008-02-23 15:24:08 -0800561struct cgroup_subsys_state *create(struct cgroup_subsys *ss,
562 struct cgroup *cgrp)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800563(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700564
565Called to create a subsystem state object for a cgroup. The
566subsystem should allocate its subsystem state object for the passed
567cgroup, returning a pointer to the new object on success or a
568negative error code. On success, the subsystem pointer should point to
569a structure of type cgroup_subsys_state (typically embedded in a
570larger subsystem-specific object), which will be initialized by the
571cgroup system. Note that this will be called at initialization to
572create the root subsystem state for this subsystem; this case can be
573identified by the passed cgroup object having a NULL parent (since
574it's the root of the hierarchy) and may be an appropriate place for
575initialization code.
576
Li Zefand19e0582008-02-23 15:24:08 -0800577void destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800578(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700579
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800580The cgroup system is about to destroy the passed cgroup; the subsystem
581should do any necessary cleanup and free its subsystem state
582object. By the time this method is called, the cgroup has already been
583unlinked from the file system and from the child list of its parent;
584cgroup->parent is still valid. (Note - can also be called for a
585newly-created cgroup if an error occurs after this subsystem's
586create() method has been called for the new cgroup).
Paul Menageddbcc7e2007-10-18 23:39:30 -0700587
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700588int pre_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp);
Li Zefand19e0582008-02-23 15:24:08 -0800589
590Called before checking the reference count on each subsystem. This may
591be useful for subsystems which have some extra references even if
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700592there are not tasks in the cgroup. If pre_destroy() returns error code,
593rmdir() will fail with it. From this behavior, pre_destroy() can be
594called multiple times against a cgroup.
Li Zefand19e0582008-02-23 15:24:08 -0800595
596int can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumf780bdb2011-05-26 16:25:19 -0700597 struct task_struct *task)
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800598(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700599
600Called prior to moving a task into a cgroup; if the subsystem
601returns an error, this will abort the attach operation. If a NULL
602task is passed, then a successful result indicates that *any*
603unspecified task can be moved into the cgroup. Note that this isn't
604called on a fork. If this method returns 0 (success) then this should
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800605remain valid while the caller holds cgroup_mutex and it is ensured that either
Ben Blumf780bdb2011-05-26 16:25:19 -0700606attach() or cancel_attach() will be called in future.
607
608int can_attach_task(struct cgroup *cgrp, struct task_struct *tsk);
609(cgroup_mutex held by caller)
610
611As can_attach, but for operations that must be run once per task to be
612attached (possibly many when using cgroup_attach_proc). Called after
613can_attach.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700614
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800615void cancel_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
616 struct task_struct *task, bool threadgroup)
617(cgroup_mutex held by caller)
618
619Called when a task attach operation has failed after can_attach() has succeeded.
620A subsystem whose can_attach() has some side-effects should provide this
Thomas Weber88393162010-03-16 11:47:56 +0100621function, so that the subsystem can implement a rollback. If not, not necessary.
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800622This will be called only about subsystems whose can_attach() operation have
623succeeded.
624
Ben Blumf780bdb2011-05-26 16:25:19 -0700625void pre_attach(struct cgroup *cgrp);
626(cgroup_mutex held by caller)
627
628For any non-per-thread attachment work that needs to happen before
629attach_task. Needed by cpuset.
630
Li Zefand19e0582008-02-23 15:24:08 -0800631void attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumf780bdb2011-05-26 16:25:19 -0700632 struct cgroup *old_cgrp, struct task_struct *task)
Li Zefan18e7f1f2009-01-07 18:07:32 -0800633(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700634
635Called after the task has been attached to the cgroup, to allow any
636post-attachment activity that requires memory allocations or blocking.
Ben Blumf780bdb2011-05-26 16:25:19 -0700637
638void attach_task(struct cgroup *cgrp, struct task_struct *tsk);
639(cgroup_mutex held by caller)
640
641As attach, but for operations that must be run once per task to be attached,
642like can_attach_task. Called before attach. Currently does not support any
Ben Blumbe367d02009-09-23 15:56:31 -0700643subsystem that might need the old_cgrp for every thread in the group.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700644
645void fork(struct cgroup_subsy *ss, struct task_struct *task)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700646
Li Zefane8d55fd2008-04-29 01:00:13 -0700647Called when a task is forked into a cgroup.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700648
649void exit(struct cgroup_subsys *ss, struct task_struct *task)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700650
Li Zefand19e0582008-02-23 15:24:08 -0800651Called during task exit.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700652
Li Zefand19e0582008-02-23 15:24:08 -0800653int populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
Li Zefan18e7f1f2009-01-07 18:07:32 -0800654(cgroup_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700655
656Called after creation of a cgroup to allow a subsystem to populate
657the cgroup directory with file entries. The subsystem should make
658calls to cgroup_add_file() with objects of type cftype (see
659include/linux/cgroup.h for details). Note that although this
660method can return an error code, the error code is currently not
661always handled well.
662
Li Zefand19e0582008-02-23 15:24:08 -0800663void post_clone(struct cgroup_subsys *ss, struct cgroup *cgrp)
Li Zefan18e7f1f2009-01-07 18:07:32 -0800664(cgroup_mutex held by caller)
Paul Menage697f4162007-10-18 23:39:34 -0700665
Daniel Lezcanoa77aea92011-05-26 16:25:23 -0700666Called during cgroup_create() to do any parameter
Paul Menage697f4162007-10-18 23:39:34 -0700667initialization which might be required before a task could attach. For
668example in cpusets, no task may attach before 'cpus' and 'mems' are set
669up.
670
Paul Menageddbcc7e2007-10-18 23:39:30 -0700671void bind(struct cgroup_subsys *ss, struct cgroup *root)
Paul Menage999cd8a2009-01-07 18:08:36 -0800672(cgroup_mutex and ss->hierarchy_mutex held by caller)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700673
674Called when a cgroup subsystem is rebound to a different hierarchy
675and root cgroup. Currently this will only involve movement between
676the default hierarchy (which never has sub-cgroups) and a hierarchy
677that is being created/destroyed (and hence has no sub-cgroups).
678
6794. Questions
680============
681
682Q: what's up with this '/bin/echo' ?
683A: bash's builtin 'echo' command does not check calls to write() against
684 errors. If you use it in the cgroup file system, you won't be
685 able to tell whether a command succeeded or failed.
686
687Q: When I attach processes, only the first of the line gets really attached !
688A: We can only return one error code per call to write(). So you should also
689 put only ONE pid.
690