aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/cgroups.txt
diff options
context:
space:
mode:
authorPaul Menage <menage@google.com>2007-10-19 02:39:30 -0400
committerLinus Torvalds <torvalds@woody.linux-foundation.org>2007-10-19 14:53:36 -0400
commitddbcc7e8e50aefe467c01cac3dec71f118cd8ac2 (patch)
tree0881a031e669582f819d572339e955b04abfc3d2 /Documentation/cgroups.txt
parent55a230aae650157720becc09cadb7d10efbf5013 (diff)
Task Control Groups: basic task cgroup framework
Generic Process Control Groups -------------------------- There have recently been various proposals floating around for resource management/accounting and other task grouping subsystems in the kernel, including ResGroups, User BeanCounters, NSProxy cgroups, and others. These all need the basic abstraction of being able to group together multiple processes in an aggregate, in order to track/limit the resources permitted to those processes, or control other behaviour of the processes, and all implement this grouping in different ways. This patchset provides a framework for tracking and grouping processes into arbitrary "cgroups" and assigning arbitrary state to those groupings, in order to control the behaviour of the cgroup as an aggregate. The intention is that the various resource management and virtualization/cgroup efforts can also become task cgroup clients, with the result that: - the userspace APIs are (somewhat) normalised - it's easier to test e.g. the ResGroups CPU controller in conjunction with the BeanCounters memory controller, or use either of them as the resource-control portion of a virtual server system. - the additional kernel footprint of any of the competing resource management systems is substantially reduced, since it doesn't need to provide process grouping/containment, hence improving their chances of getting into the kernel This patch: Add the main task cgroups framework - the cgroup filesystem, and the basic structures for tracking membership and associating subsystem state objects to tasks. Signed-off-by: Paul Menage <menage@google.com> Cc: Serge E. Hallyn <serue@us.ibm.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Dave Hansen <haveblue@us.ibm.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Paul Jackson <pj@sgi.com> Cc: Kirill Korotaev <dev@openvz.org> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com> Cc: Cedric Le Goater <clg@fr.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'Documentation/cgroups.txt')
-rw-r--r--Documentation/cgroups.txt526
1 files changed, 526 insertions, 0 deletions
diff --git a/Documentation/cgroups.txt b/Documentation/cgroups.txt
new file mode 100644
index 000000000000..4717887fd75d
--- /dev/null
+++ b/Documentation/cgroups.txt
@@ -0,0 +1,526 @@
1 CGROUPS
2 -------
3
4Written by Paul Menage <menage@google.com> based on Documentation/cpusets.txt
5
6Original copyright statements from cpusets.txt:
7Portions Copyright (C) 2004 BULL SA.
8Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
9Modified by Paul Jackson <pj@sgi.com>
10Modified by Christoph Lameter <clameter@sgi.com>
11
12CONTENTS:
13=========
14
151. Control Groups
16 1.1 What are cgroups ?
17 1.2 Why are cgroups needed ?
18 1.3 How are cgroups implemented ?
19 1.4 What does notify_on_release do ?
20 1.5 How do I use cgroups ?
212. Usage Examples and Syntax
22 2.1 Basic Usage
23 2.2 Attaching processes
243. Kernel API
25 3.1 Overview
26 3.2 Synchronization
27 3.3 Subsystem API
284. Questions
29
301. Control Groups
31==========
32
331.1 What are cgroups ?
34----------------------
35
36Control Groups provide a mechanism for aggregating/partitioning sets of
37tasks, and all their future children, into hierarchical groups with
38specialized behaviour.
39
40Definitions:
41
42A *cgroup* associates a set of tasks with a set of parameters for one
43or more subsystems.
44
45A *subsystem* is a module that makes use of the task grouping
46facilities provided by cgroups to treat groups of tasks in
47particular ways. A subsystem is typically a "resource controller" that
48schedules a resource or applies per-cgroup limits, but it may be
49anything that wants to act on a group of processes, e.g. a
50virtualization subsystem.
51
52A *hierarchy* is a set of cgroups arranged in a tree, such that
53every task in the system is in exactly one of the cgroups in the
54hierarchy, and a set of subsystems; each subsystem has system-specific
55state attached to each cgroup in the hierarchy. Each hierarchy has
56an instance of the cgroup virtual filesystem associated with it.
57
58At any one time there may be multiple active hierachies of task
59cgroups. Each hierarchy is a partition of all tasks in the system.
60
61User level code may create and destroy cgroups by name in an
62instance of the cgroup virtual file system, specify and query to
63which cgroup a task is assigned, and list the task pids assigned to
64a cgroup. Those creations and assignments only affect the hierarchy
65associated with that instance of the cgroup file system.
66
67On their own, the only use for cgroups is for simple job
68tracking. The intention is that other subsystems hook into the generic
69cgroup support to provide new attributes for cgroups, such as
70accounting/limiting the resources which processes in a cgroup can
71access. For example, cpusets (see Documentation/cpusets.txt) allows
72you to associate a set of CPUs and a set of memory nodes with the
73tasks in each cgroup.
74
751.2 Why are cgroups needed ?
76----------------------------
77
78There are multiple efforts to provide process aggregations in the
79Linux kernel, mainly for resource tracking purposes. Such efforts
80include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server
81namespaces. These all require the basic notion of a
82grouping/partitioning of processes, with newly forked processes ending
83in the same group (cgroup) as their parent process.
84
85The kernel cgroup patch provides the minimum essential kernel
86mechanisms required to efficiently implement such groups. It has
87minimal impact on the system fast paths, and provides hooks for
88specific subsystems such as cpusets to provide additional behaviour as
89desired.
90
91Multiple hierarchy support is provided to allow for situations where
92the division of tasks into cgroups is distinctly different for
93different subsystems - having parallel hierarchies allows each
94hierarchy to be a natural division of tasks, without having to handle
95complex combinations of tasks that would be present if several
96unrelated subsystems needed to be forced into the same tree of
97cgroups.
98
99At one extreme, each resource controller or subsystem could be in a
100separate hierarchy; at the other extreme, all subsystems
101would be attached to the same hierarchy.
102
103As an example of a scenario (originally proposed by vatsa@in.ibm.com)
104that can benefit from multiple hierarchies, consider a large
105university server with various users - students, professors, system
106tasks etc. The resource planning for this server could be along the
107following lines:
108
109 CPU : Top cpuset
110 / \
111 CPUSet1 CPUSet2
112 | |
113 (Profs) (Students)
114
115 In addition (system tasks) are attached to topcpuset (so
116 that they can run anywhere) with a limit of 20%
117
118 Memory : Professors (50%), students (30%), system (20%)
119
120 Disk : Prof (50%), students (30%), system (20%)
121
122 Network : WWW browsing (20%), Network File System (60%), others (20%)
123 / \
124 Prof (15%) students (5%)
125
126Browsers like firefox/lynx go into the WWW network class, while (k)nfsd go
127into NFS network class.
128
129At the same time firefox/lynx will share an appropriate CPU/Memory class
130depending on who launched it (prof/student).
131
132With the ability to classify tasks differently for different resources
133(by putting those resource subsystems in different hierarchies) then
134the admin can easily set up a script which receives exec notifications
135and depending on who is launching the browser he can
136
137 # echo browser_pid > /mnt/<restype>/<userclass>/tasks
138
139With only a single hierarchy, he now would potentially have to create
140a separate cgroup for every browser launched and associate it with
141approp network and other resource class. This may lead to
142proliferation of such cgroups.
143
144Also lets say that the administrator would like to give enhanced network
145access temporarily to a student's browser (since it is night and the user
146wants to do online gaming :) OR give one of the students simulation
147apps enhanced CPU power,
148
149With ability to write pids directly to resource classes, its just a
150matter of :
151
152 # echo pid > /mnt/network/<new_class>/tasks
153 (after some time)
154 # echo pid > /mnt/network/<orig_class>/tasks
155
156Without this ability, he would have to split the cgroup into
157multiple separate ones and then associate the new cgroups with the
158new resource classes.
159
160
161
1621.3 How are cgroups implemented ?
163---------------------------------
164
165Control Groups extends the kernel as follows:
166
167 - Each task in the system has a reference-counted pointer to a
168 css_set.
169
170 - A css_set contains a set of reference-counted pointers to
171 cgroup_subsys_state objects, one for each cgroup subsystem
172 registered in the system. There is no direct link from a task to
173 the cgroup of which it's a member in each hierarchy, but this
174 can be determined by following pointers through the
175 cgroup_subsys_state objects. This is because accessing the
176 subsystem state is something that's expected to happen frequently
177 and in performance-critical code, whereas operations that require a
178 task's actual cgroup assignments (in particular, moving between
179 cgroups) are less common.
180
181 - A cgroup hierarchy filesystem can be mounted for browsing and
182 manipulation from user space.
183
184 - You can list all the tasks (by pid) attached to any cgroup.
185
186The implementation of cgroups requires a few, simple hooks
187into the rest of the kernel, none in performance critical paths:
188
189 - in init/main.c, to initialize the root cgroups and initial
190 css_set at system boot.
191
192 - in fork and exit, to attach and detach a task from its css_set.
193
194In addition a new file system, of type "cgroup" may be mounted, to
195enable browsing and modifying the cgroups presently known to the
196kernel. When mounting a cgroup hierarchy, you may specify a
197comma-separated list of subsystems to mount as the filesystem mount
198options. By default, mounting the cgroup filesystem attempts to
199mount a hierarchy containing all registered subsystems.
200
201If an active hierarchy with exactly the same set of subsystems already
202exists, it will be reused for the new mount. If no existing hierarchy
203matches, and any of the requested subsystems are in use in an existing
204hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy
205is activated, associated with the requested subsystems.
206
207It's not currently possible to bind a new subsystem to an active
208cgroup hierarchy, or to unbind a subsystem from an active cgroup
209hierarchy. This may be possible in future, but is fraught with nasty
210error-recovery issues.
211
212When a cgroup filesystem is unmounted, if there are any
213child cgroups created below the top-level cgroup, that hierarchy
214will remain active even though unmounted; if there are no
215child cgroups then the hierarchy will be deactivated.
216
217No new system calls are added for cgroups - all support for
218querying and modifying cgroups is via this cgroup file system.
219
220Each task under /proc has an added file named 'cgroup' displaying,
221for each active hierarchy, the subsystem names and the cgroup name
222as the path relative to the root of the cgroup file system.
223
224Each cgroup is represented by a directory in the cgroup file system
225containing the following files describing that cgroup:
226
227 - tasks: list of tasks (by pid) attached to that cgroup
228 - notify_on_release flag: run /sbin/cgroup_release_agent on exit?
229
230Other subsystems such as cpusets may add additional files in each
231cgroup dir
232
233New cgroups are created using the mkdir system call or shell
234command. The properties of a cgroup, such as its flags, are
235modified by writing to the appropriate file in that cgroups
236directory, as listed above.
237
238The named hierarchical structure of nested cgroups allows partitioning
239a large system into nested, dynamically changeable, "soft-partitions".
240
241The attachment of each task, automatically inherited at fork by any
242children of that task, to a cgroup allows organizing the work load
243on a system into related sets of tasks. A task may be re-attached to
244any other cgroup, if allowed by the permissions on the necessary
245cgroup file system directories.
246
247When a task is moved from one cgroup to another, it gets a new
248css_set pointer - if there's an already existing css_set with the
249desired collection of cgroups then that group is reused, else a new
250css_set is allocated. Note that the current implementation uses a
251linear search to locate an appropriate existing css_set, so isn't
252very efficient. A future version will use a hash table for better
253performance.
254
255The use of a Linux virtual file system (vfs) to represent the
256cgroup hierarchy provides for a familiar permission and name space
257for cgroups, with a minimum of additional kernel code.
258
2591.4 What does notify_on_release do ?
260------------------------------------
261
262*** notify_on_release is disabled in the current patch set. It will be
263*** reactivated in a future patch in a less-intrusive manner
264
265If the notify_on_release flag is enabled (1) in a cgroup, then
266whenever the last task in the cgroup leaves (exits or attaches to
267some other cgroup) and the last child cgroup of that cgroup
268is removed, then the kernel runs the command specified by the contents
269of the "release_agent" file in that hierarchy's root directory,
270supplying the pathname (relative to the mount point of the cgroup
271file system) of the abandoned cgroup. This enables automatic
272removal of abandoned cgroups. The default value of
273notify_on_release in the root cgroup at system boot is disabled
274(0). The default value of other cgroups at creation is the current
275value of their parents notify_on_release setting. The default value of
276a cgroup hierarchy's release_agent path is empty.
277
2781.5 How do I use cgroups ?
279--------------------------
280
281To start a new job that is to be contained within a cgroup, using
282the "cpuset" cgroup subsystem, the steps are something like:
283
284 1) mkdir /dev/cgroup
285 2) mount -t cgroup -ocpuset cpuset /dev/cgroup
286 3) Create the new cgroup by doing mkdir's and write's (or echo's) in
287 the /dev/cgroup virtual file system.
288 4) Start a task that will be the "founding father" of the new job.
289 5) Attach that task to the new cgroup by writing its pid to the
290 /dev/cgroup tasks file for that cgroup.
291 6) fork, exec or clone the job tasks from this founding father task.
292
293For example, the following sequence of commands will setup a cgroup
294named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
295and then start a subshell 'sh' in that cgroup:
296
297 mount -t cgroup cpuset -ocpuset /dev/cgroup
298 cd /dev/cgroup
299 mkdir Charlie
300 cd Charlie
301 /bin/echo 2-3 > cpus
302 /bin/echo 1 > mems
303 /bin/echo $$ > tasks
304 sh
305 # The subshell 'sh' is now running in cgroup Charlie
306 # The next line should display '/Charlie'
307 cat /proc/self/cgroup
308
3092. Usage Examples and Syntax
310============================
311
3122.1 Basic Usage
313---------------
314
315Creating, modifying, using the cgroups can be done through the cgroup
316virtual filesystem.
317
318To mount a cgroup hierarchy will all available subsystems, type:
319# mount -t cgroup xxx /dev/cgroup
320
321The "xxx" is not interpreted by the cgroup code, but will appear in
322/proc/mounts so may be any useful identifying string that you like.
323
324To mount a cgroup hierarchy with just the cpuset and numtasks
325subsystems, type:
326# mount -t cgroup -o cpuset,numtasks hier1 /dev/cgroup
327
328To change the set of subsystems bound to a mounted hierarchy, just
329remount with different options:
330
331# mount -o remount,cpuset,ns /dev/cgroup
332
333Note that changing the set of subsystems is currently only supported
334when the hierarchy consists of a single (root) cgroup. Supporting
335the ability to arbitrarily bind/unbind subsystems from an existing
336cgroup hierarchy is intended to be implemented in the future.
337
338Then under /dev/cgroup you can find a tree that corresponds to the
339tree of the cgroups in the system. For instance, /dev/cgroup
340is the cgroup that holds the whole system.
341
342If you want to create a new cgroup under /dev/cgroup:
343# cd /dev/cgroup
344# mkdir my_cgroup
345
346Now you want to do something with this cgroup.
347# cd my_cgroup
348
349In this directory you can find several files:
350# ls
351notify_on_release release_agent tasks
352(plus whatever files are added by the attached subsystems)
353
354Now attach your shell to this cgroup:
355# /bin/echo $$ > tasks
356
357You can also create cgroups inside your cgroup by using mkdir in this
358directory.
359# mkdir my_sub_cs
360
361To remove a cgroup, just use rmdir:
362# rmdir my_sub_cs
363
364This will fail if the cgroup is in use (has cgroups inside, or
365has processes attached, or is held alive by other subsystem-specific
366reference).
367
3682.2 Attaching processes
369-----------------------
370
371# /bin/echo PID > tasks
372
373Note that it is PID, not PIDs. You can only attach ONE task at a time.
374If you have several tasks to attach, you have to do it one after another:
375
376# /bin/echo PID1 > tasks
377# /bin/echo PID2 > tasks
378 ...
379# /bin/echo PIDn > tasks
380
3813. Kernel API
382=============
383
3843.1 Overview
385------------
386
387Each kernel subsystem that wants to hook into the generic cgroup
388system needs to create a cgroup_subsys object. This contains
389various methods, which are callbacks from the cgroup system, along
390with a subsystem id which will be assigned by the cgroup system.
391
392Other fields in the cgroup_subsys object include:
393
394- subsys_id: a unique array index for the subsystem, indicating which
395 entry in cgroup->subsys[] this subsystem should be
396 managing. Initialized by cgroup_register_subsys(); prior to this
397 it should be initialized to -1
398
399- hierarchy: an index indicating which hierarchy, if any, this
400 subsystem is currently attached to. If this is -1, then the
401 subsystem is not attached to any hierarchy, and all tasks should be
402 considered to be members of the subsystem's top_cgroup. It should
403 be initialized to -1.
404
405- name: should be initialized to a unique subsystem name prior to
406 calling cgroup_register_subsystem. Should be no longer than
407 MAX_CGROUP_TYPE_NAMELEN
408
409Each cgroup object created by the system has an array of pointers,
410indexed by subsystem id; this pointer is entirely managed by the
411subsystem; the generic cgroup code will never touch this pointer.
412
4133.2 Synchronization
414-------------------
415
416There is a global mutex, cgroup_mutex, used by the cgroup
417system. This should be taken by anything that wants to modify a
418cgroup. It may also be taken to prevent cgroups from being
419modified, but more specific locks may be more appropriate in that
420situation.
421
422See kernel/cgroup.c for more details.
423
424Subsystems can take/release the cgroup_mutex via the functions
425cgroup_lock()/cgroup_unlock(), and can
426take/release the callback_mutex via the functions
427cgroup_lock()/cgroup_unlock().
428
429Accessing a task's cgroup pointer may be done in the following ways:
430- while holding cgroup_mutex
431- while holding the task's alloc_lock (via task_lock())
432- inside an rcu_read_lock() section via rcu_dereference()
433
4343.3 Subsystem API
435--------------------------
436
437Each subsystem should:
438
439- add an entry in linux/cgroup_subsys.h
440- define a cgroup_subsys object called <name>_subsys
441
442Each subsystem may export the following methods. The only mandatory
443methods are create/destroy. Any others that are null are presumed to
444be successful no-ops.
445
446struct cgroup_subsys_state *create(struct cgroup *cont)
447LL=cgroup_mutex
448
449Called to create a subsystem state object for a cgroup. The
450subsystem should allocate its subsystem state object for the passed
451cgroup, returning a pointer to the new object on success or a
452negative error code. On success, the subsystem pointer should point to
453a structure of type cgroup_subsys_state (typically embedded in a
454larger subsystem-specific object), which will be initialized by the
455cgroup system. Note that this will be called at initialization to
456create the root subsystem state for this subsystem; this case can be
457identified by the passed cgroup object having a NULL parent (since
458it's the root of the hierarchy) and may be an appropriate place for
459initialization code.
460
461void destroy(struct cgroup *cont)
462LL=cgroup_mutex
463
464The cgroup system is about to destroy the passed cgroup; the
465subsystem should do any necessary cleanup
466
467int can_attach(struct cgroup_subsys *ss, struct cgroup *cont,
468 struct task_struct *task)
469LL=cgroup_mutex
470
471Called prior to moving a task into a cgroup; if the subsystem
472returns an error, this will abort the attach operation. If a NULL
473task is passed, then a successful result indicates that *any*
474unspecified task can be moved into the cgroup. Note that this isn't
475called on a fork. If this method returns 0 (success) then this should
476remain valid while the caller holds cgroup_mutex.
477
478void attach(struct cgroup_subsys *ss, struct cgroup *cont,
479 struct cgroup *old_cont, struct task_struct *task)
480LL=cgroup_mutex
481
482
483Called after the task has been attached to the cgroup, to allow any
484post-attachment activity that requires memory allocations or blocking.
485
486void fork(struct cgroup_subsy *ss, struct task_struct *task)
487LL=callback_mutex, maybe read_lock(tasklist_lock)
488
489Called when a task is forked into a cgroup. Also called during
490registration for all existing tasks.
491
492void exit(struct cgroup_subsys *ss, struct task_struct *task)
493LL=callback_mutex
494
495Called during task exit
496
497int populate(struct cgroup_subsys *ss, struct cgroup *cont)
498LL=none
499
500Called after creation of a cgroup to allow a subsystem to populate
501the cgroup directory with file entries. The subsystem should make
502calls to cgroup_add_file() with objects of type cftype (see
503include/linux/cgroup.h for details). Note that although this
504method can return an error code, the error code is currently not
505always handled well.
506
507void bind(struct cgroup_subsys *ss, struct cgroup *root)
508LL=callback_mutex
509
510Called when a cgroup subsystem is rebound to a different hierarchy
511and root cgroup. Currently this will only involve movement between
512the default hierarchy (which never has sub-cgroups) and a hierarchy
513that is being created/destroyed (and hence has no sub-cgroups).
514
5154. Questions
516============
517
518Q: what's up with this '/bin/echo' ?
519A: bash's builtin 'echo' command does not check calls to write() against
520 errors. If you use it in the cgroup file system, you won't be
521 able to tell whether a command succeeded or failed.
522
523Q: When I attach processes, only the first of the line gets really attached !
524A: We can only return one error code per call to write(). So you should also
525 put only ONE pid.
526