Commit | Line | Data |
---|---|---|
ddbcc7e8 PM |
1 | CGROUPS |
2 | ------- | |
3 | ||
45ce80fb LZ |
4 | Written by Paul Menage <menage@google.com> based on |
5 | Documentation/cgroups/cpusets.txt | |
ddbcc7e8 PM |
6 | |
7 | Original copyright statements from cpusets.txt: | |
8 | Portions Copyright (C) 2004 BULL SA. | |
9 | Portions Copyright (c) 2004-2006 Silicon Graphics, Inc. | |
10 | Modified by Paul Jackson <pj@sgi.com> | |
93e205a7 | 11 | Modified by Christoph Lameter <cl@linux.com> |
ddbcc7e8 PM |
12 | |
13 | CONTENTS: | |
14 | ========= | |
15 | ||
16 | 1. 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 ? | |
97978e6d DL |
21 | 1.5 What does clone_children do ? |
22 | 1.6 How do I use cgroups ? | |
ddbcc7e8 PM |
23 | 2. Usage Examples and Syntax |
24 | 2.1 Basic Usage | |
25 | 2.2 Attaching processes | |
8ca712ea | 26 | 2.3 Mounting hierarchies by name |
ddbcc7e8 PM |
27 | 3. Kernel API |
28 | 3.1 Overview | |
29 | 3.2 Synchronization | |
30 | 3.3 Subsystem API | |
19ec2567 AR |
31 | 4. Extended attributes usage |
32 | 5. Questions | |
ddbcc7e8 PM |
33 | |
34 | 1. Control Groups | |
d19e0583 | 35 | ================= |
ddbcc7e8 PM |
36 | |
37 | 1.1 What are cgroups ? | |
38 | ---------------------- | |
39 | ||
40 | Control Groups provide a mechanism for aggregating/partitioning sets of | |
41 | tasks, and all their future children, into hierarchical groups with | |
42 | specialized behaviour. | |
43 | ||
44 | Definitions: | |
45 | ||
46 | A *cgroup* associates a set of tasks with a set of parameters for one | |
47 | or more subsystems. | |
48 | ||
49 | A *subsystem* is a module that makes use of the task grouping | |
50 | facilities provided by cgroups to treat groups of tasks in | |
51 | particular ways. A subsystem is typically a "resource controller" that | |
52 | schedules a resource or applies per-cgroup limits, but it may be | |
53 | anything that wants to act on a group of processes, e.g. a | |
54 | virtualization subsystem. | |
55 | ||
56 | A *hierarchy* is a set of cgroups arranged in a tree, such that | |
57 | every task in the system is in exactly one of the cgroups in the | |
58 | hierarchy, and a set of subsystems; each subsystem has system-specific | |
59 | state attached to each cgroup in the hierarchy. Each hierarchy has | |
60 | an instance of the cgroup virtual filesystem associated with it. | |
61 | ||
caa790ba | 62 | At any one time there may be multiple active hierarchies of task |
ddbcc7e8 PM |
63 | cgroups. Each hierarchy is a partition of all tasks in the system. |
64 | ||
83b061fc | 65 | User-level code may create and destroy cgroups by name in an |
ddbcc7e8 | 66 | instance of the cgroup virtual file system, specify and query to |
83b061fc | 67 | which cgroup a task is assigned, and list the task PIDs assigned to |
ddbcc7e8 PM |
68 | a cgroup. Those creations and assignments only affect the hierarchy |
69 | associated with that instance of the cgroup file system. | |
70 | ||
71 | On their own, the only use for cgroups is for simple job | |
72 | tracking. The intention is that other subsystems hook into the generic | |
73 | cgroup support to provide new attributes for cgroups, such as | |
74 | accounting/limiting the resources which processes in a cgroup can | |
83b061fc | 75 | access. For example, cpusets (see Documentation/cgroups/cpusets.txt) allow |
ddbcc7e8 PM |
76 | you to associate a set of CPUs and a set of memory nodes with the |
77 | tasks in each cgroup. | |
78 | ||
79 | 1.2 Why are cgroups needed ? | |
80 | ---------------------------- | |
81 | ||
82 | There are multiple efforts to provide process aggregations in the | |
83b061fc | 83 | Linux kernel, mainly for resource-tracking purposes. Such efforts |
ddbcc7e8 PM |
84 | include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server |
85 | namespaces. These all require the basic notion of a | |
86 | grouping/partitioning of processes, with newly forked processes ending | |
83b061fc | 87 | up in the same group (cgroup) as their parent process. |
ddbcc7e8 PM |
88 | |
89 | The kernel cgroup patch provides the minimum essential kernel | |
90 | mechanisms required to efficiently implement such groups. It has | |
91 | minimal impact on the system fast paths, and provides hooks for | |
92 | specific subsystems such as cpusets to provide additional behaviour as | |
93 | desired. | |
94 | ||
95 | Multiple hierarchy support is provided to allow for situations where | |
96 | the division of tasks into cgroups is distinctly different for | |
97 | different subsystems - having parallel hierarchies allows each | |
98 | hierarchy to be a natural division of tasks, without having to handle | |
99 | complex combinations of tasks that would be present if several | |
100 | unrelated subsystems needed to be forced into the same tree of | |
101 | cgroups. | |
102 | ||
103 | At one extreme, each resource controller or subsystem could be in a | |
104 | separate hierarchy; at the other extreme, all subsystems | |
105 | would be attached to the same hierarchy. | |
106 | ||
107 | As an example of a scenario (originally proposed by vatsa@in.ibm.com) | |
108 | that can benefit from multiple hierarchies, consider a large | |
109 | university server with various users - students, professors, system | |
110 | tasks etc. The resource planning for this server could be along the | |
111 | following lines: | |
112 | ||
6ad85239 | 113 | CPU : "Top cpuset" |
ddbcc7e8 PM |
114 | / \ |
115 | CPUSet1 CPUSet2 | |
6ad85239 GL |
116 | | | |
117 | (Professors) (Students) | |
ddbcc7e8 PM |
118 | |
119 | In addition (system tasks) are attached to topcpuset (so | |
120 | that they can run anywhere) with a limit of 20% | |
121 | ||
6ad85239 | 122 | Memory : Professors (50%), Students (30%), system (20%) |
ddbcc7e8 | 123 | |
6ad85239 | 124 | Disk : Professors (50%), Students (30%), system (20%) |
ddbcc7e8 PM |
125 | |
126 | Network : WWW browsing (20%), Network File System (60%), others (20%) | |
127 | / \ | |
6ad85239 | 128 | Professors (15%) students (5%) |
ddbcc7e8 | 129 | |
83b061fc MK |
130 | Browsers like Firefox/Lynx go into the WWW network class, while (k)nfsd goes |
131 | into the NFS network class. | |
ddbcc7e8 | 132 | |
caa790ba | 133 | At the same time Firefox/Lynx will share an appropriate CPU/Memory class |
ddbcc7e8 PM |
134 | depending on who launched it (prof/student). |
135 | ||
136 | With the ability to classify tasks differently for different resources | |
83b061fc | 137 | (by putting those resource subsystems in different hierarchies), |
ddbcc7e8 PM |
138 | the admin can easily set up a script which receives exec notifications |
139 | and depending on who is launching the browser he can | |
140 | ||
f6e07d38 | 141 | # echo browser_pid > /sys/fs/cgroup/<restype>/<userclass>/tasks |
ddbcc7e8 PM |
142 | |
143 | With only a single hierarchy, he now would potentially have to create | |
144 | a separate cgroup for every browser launched and associate it with | |
67de0162 | 145 | appropriate network and other resource class. This may lead to |
ddbcc7e8 PM |
146 | proliferation of such cgroups. |
147 | ||
83b061fc | 148 | Also let's say that the administrator would like to give enhanced network |
ddbcc7e8 | 149 | access temporarily to a student's browser (since it is night and the user |
83b061fc MK |
150 | wants to do online gaming :)) OR give one of the student's simulation |
151 | apps enhanced CPU power. | |
ddbcc7e8 | 152 | |
83b061fc MK |
153 | With ability to write PIDs directly to resource classes, it's just a |
154 | matter of: | |
ddbcc7e8 | 155 | |
f6e07d38 | 156 | # echo pid > /sys/fs/cgroup/network/<new_class>/tasks |
ddbcc7e8 | 157 | (after some time) |
f6e07d38 | 158 | # echo pid > /sys/fs/cgroup/network/<orig_class>/tasks |
ddbcc7e8 | 159 | |
83b061fc | 160 | Without this ability, the administrator would have to split the cgroup into |
ddbcc7e8 PM |
161 | multiple separate ones and then associate the new cgroups with the |
162 | new resource classes. | |
163 | ||
164 | ||
165 | ||
166 | 1.3 How are cgroups implemented ? | |
167 | --------------------------------- | |
168 | ||
169 | Control 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 | |
817929ec PM |
183 | 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. | |
ddbcc7e8 | 186 | |
83b061fc | 187 | - A cgroup hierarchy filesystem can be mounted for browsing and |
ddbcc7e8 PM |
188 | manipulation from user space. |
189 | ||
83b061fc | 190 | - You can list all the tasks (by PID) attached to any cgroup. |
ddbcc7e8 PM |
191 | |
192 | The implementation of cgroups requires a few, simple hooks | |
83b061fc | 193 | into the rest of the kernel, none in performance-critical paths: |
ddbcc7e8 PM |
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 | ||
83b061fc | 200 | In addition, a new file system of type "cgroup" may be mounted, to |
ddbcc7e8 PM |
201 | enable browsing and modifying the cgroups presently known to the |
202 | kernel. When mounting a cgroup hierarchy, you may specify a | |
203 | comma-separated list of subsystems to mount as the filesystem mount | |
204 | options. By default, mounting the cgroup filesystem attempts to | |
205 | mount a hierarchy containing all registered subsystems. | |
206 | ||
207 | If an active hierarchy with exactly the same set of subsystems already | |
208 | exists, it will be reused for the new mount. If no existing hierarchy | |
209 | matches, and any of the requested subsystems are in use in an existing | |
210 | hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy | |
211 | is activated, associated with the requested subsystems. | |
212 | ||
26d5bbe5 TH |
213 | It's not currently possible to bind a new subsystem to an active |
214 | cgroup hierarchy, or to unbind a subsystem from an active cgroup | |
215 | hierarchy. This may be possible in future, but is fraught with nasty | |
216 | error-recovery issues. | |
ddbcc7e8 PM |
217 | |
218 | When a cgroup filesystem is unmounted, if there are any | |
219 | child cgroups created below the top-level cgroup, that hierarchy | |
220 | will remain active even though unmounted; if there are no | |
221 | child cgroups then the hierarchy will be deactivated. | |
222 | ||
223 | No new system calls are added for cgroups - all support for | |
224 | querying and modifying cgroups is via this cgroup file system. | |
225 | ||
226 | Each task under /proc has an added file named 'cgroup' displaying, | |
227 | for each active hierarchy, the subsystem names and the cgroup name | |
228 | as the path relative to the root of the cgroup file system. | |
229 | ||
230 | Each cgroup is represented by a directory in the cgroup file system | |
231 | containing the following files describing that cgroup: | |
232 | ||
83b061fc MK |
233 | - 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 | |
7823da36 | 235 | moves the thread into this cgroup. |
83b061fc MK |
236 | - cgroup.procs: list of thread group IDs in the cgroup. This list is |
237 | not guaranteed to be sorted or free of duplicate TGIDs, and userspace | |
7823da36 | 238 | should sort/uniquify the list if this property is required. |
83b061fc | 239 | Writing a thread group ID into this file moves all threads in that |
74a1166d | 240 | group into this cgroup. |
d19e0583 LZ |
241 | - 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) | |
ddbcc7e8 PM |
244 | |
245 | Other subsystems such as cpusets may add additional files in each | |
d19e0583 | 246 | cgroup dir. |
ddbcc7e8 PM |
247 | |
248 | New cgroups are created using the mkdir system call or shell | |
249 | command. The properties of a cgroup, such as its flags, are | |
250 | modified by writing to the appropriate file in that cgroups | |
251 | directory, as listed above. | |
252 | ||
253 | The named hierarchical structure of nested cgroups allows partitioning | |
254 | a large system into nested, dynamically changeable, "soft-partitions". | |
255 | ||
256 | The attachment of each task, automatically inherited at fork by any | |
257 | children of that task, to a cgroup allows organizing the work load | |
258 | on a system into related sets of tasks. A task may be re-attached to | |
259 | any other cgroup, if allowed by the permissions on the necessary | |
260 | cgroup file system directories. | |
261 | ||
262 | When a task is moved from one cgroup to another, it gets a new | |
263 | css_set pointer - if there's an already existing css_set with the | |
83b061fc | 264 | desired collection of cgroups then that group is reused, otherwise a new |
b851ee79 LZ |
265 | css_set is allocated. The appropriate existing css_set is located by |
266 | looking into a hash table. | |
ddbcc7e8 | 267 | |
817929ec PM |
268 | To allow access from a cgroup to the css_sets (and hence tasks) |
269 | that comprise it, a set of cg_cgroup_link objects form a lattice; | |
270 | each cg_cgroup_link is linked into a list of cg_cgroup_links for | |
d19e0583 | 271 | a single cgroup on its cgrp_link_list field, and a list of |
817929ec PM |
272 | cg_cgroup_links for a single css_set on its cg_link_list. |
273 | ||
274 | Thus the set of tasks in a cgroup can be listed by iterating over | |
275 | each css_set that references the cgroup, and sub-iterating over | |
276 | each css_set's task set. | |
277 | ||
ddbcc7e8 PM |
278 | The use of a Linux virtual file system (vfs) to represent the |
279 | cgroup hierarchy provides for a familiar permission and name space | |
280 | for cgroups, with a minimum of additional kernel code. | |
281 | ||
282 | 1.4 What does notify_on_release do ? | |
283 | ------------------------------------ | |
284 | ||
ddbcc7e8 PM |
285 | If the notify_on_release flag is enabled (1) in a cgroup, then |
286 | whenever the last task in the cgroup leaves (exits or attaches to | |
287 | some other cgroup) and the last child cgroup of that cgroup | |
288 | is removed, then the kernel runs the command specified by the contents | |
289 | of the "release_agent" file in that hierarchy's root directory, | |
290 | supplying the pathname (relative to the mount point of the cgroup | |
291 | file system) of the abandoned cgroup. This enables automatic | |
292 | removal of abandoned cgroups. The default value of | |
293 | notify_on_release in the root cgroup at system boot is disabled | |
294 | (0). The default value of other cgroups at creation is the current | |
83b061fc | 295 | value of their parents' notify_on_release settings. The default value of |
ddbcc7e8 PM |
296 | a cgroup hierarchy's release_agent path is empty. |
297 | ||
97978e6d DL |
298 | 1.5 What does clone_children do ? |
299 | --------------------------------- | |
300 | ||
2260e7fc TH |
301 | This flag only affects the cpuset controller. If the clone_children |
302 | flag is enabled (1) in a cgroup, a new cpuset cgroup will copy its | |
303 | configuration from the parent during initialization. | |
97978e6d DL |
304 | |
305 | 1.6 How do I use cgroups ? | |
ddbcc7e8 PM |
306 | -------------------------- |
307 | ||
308 | To start a new job that is to be contained within a cgroup, using | |
309 | the "cpuset" cgroup subsystem, the steps are something like: | |
310 | ||
f6e07d38 JS |
311 | 1) mount -t tmpfs cgroup_root /sys/fs/cgroup |
312 | 2) mkdir /sys/fs/cgroup/cpuset | |
313 | 3) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset | |
314 | 4) Create the new cgroup by doing mkdir's and write's (or echo's) in | |
845502d2 | 315 | the /sys/fs/cgroup/cpuset virtual file system. |
f6e07d38 | 316 | 5) Start a task that will be the "founding father" of the new job. |
83b061fc | 317 | 6) Attach that task to the new cgroup by writing its PID to the |
845502d2 | 318 | /sys/fs/cgroup/cpuset tasks file for that cgroup. |
f6e07d38 | 319 | 7) fork, exec or clone the job tasks from this founding father task. |
ddbcc7e8 PM |
320 | |
321 | For example, the following sequence of commands will setup a cgroup | |
322 | named "Charlie", containing just CPUs 2 and 3, and Memory Node 1, | |
323 | and then start a subshell 'sh' in that cgroup: | |
324 | ||
f6e07d38 JS |
325 | mount -t tmpfs cgroup_root /sys/fs/cgroup |
326 | mkdir /sys/fs/cgroup/cpuset | |
327 | mount -t cgroup cpuset -ocpuset /sys/fs/cgroup/cpuset | |
328 | cd /sys/fs/cgroup/cpuset | |
ddbcc7e8 PM |
329 | mkdir Charlie |
330 | cd Charlie | |
0f146a76 DG |
331 | /bin/echo 2-3 > cpuset.cpus |
332 | /bin/echo 1 > cpuset.mems | |
ddbcc7e8 PM |
333 | /bin/echo $$ > tasks |
334 | sh | |
335 | # The subshell 'sh' is now running in cgroup Charlie | |
336 | # The next line should display '/Charlie' | |
337 | cat /proc/self/cgroup | |
338 | ||
339 | 2. Usage Examples and Syntax | |
340 | ============================ | |
341 | ||
342 | 2.1 Basic Usage | |
343 | --------------- | |
344 | ||
83b061fc | 345 | Creating, modifying, using cgroups can be done through the cgroup |
ddbcc7e8 PM |
346 | virtual filesystem. |
347 | ||
caa790ba | 348 | To mount a cgroup hierarchy with all available subsystems, type: |
f6e07d38 | 349 | # mount -t cgroup xxx /sys/fs/cgroup |
ddbcc7e8 PM |
350 | |
351 | The "xxx" is not interpreted by the cgroup code, but will appear in | |
352 | /proc/mounts so may be any useful identifying string that you like. | |
353 | ||
bb6405ea EM |
354 | Note: Some subsystems do not work without some user input first. For instance, |
355 | if cpusets are enabled the user will have to populate the cpus and mems files | |
356 | for each new cgroup created before that group can be used. | |
357 | ||
f6e07d38 JS |
358 | As explained in section `1.2 Why are cgroups needed?' you should create |
359 | different hierarchies of cgroups for each single resource or group of | |
360 | resources you want to control. Therefore, you should mount a tmpfs on | |
361 | /sys/fs/cgroup and create directories for each cgroup resource or resource | |
362 | group. | |
363 | ||
364 | # mount -t tmpfs cgroup_root /sys/fs/cgroup | |
365 | # mkdir /sys/fs/cgroup/rg1 | |
366 | ||
595f4b69 | 367 | To mount a cgroup hierarchy with just the cpuset and memory |
ddbcc7e8 | 368 | subsystems, type: |
f6e07d38 | 369 | # mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1 |
ddbcc7e8 | 370 | |
9a8054aa DW |
371 | While remounting cgroups is currently supported, it is not recommend |
372 | to use it. Remounting allows changing bound subsystems and | |
373 | release_agent. Rebinding is hardly useful as it only works when the | |
374 | hierarchy is empty and release_agent itself should be replaced with | |
375 | conventional fsnotify. The support for remounting will be removed in | |
376 | the future. | |
b6719ec1 LZ |
377 | |
378 | To Specify a hierarchy's release_agent: | |
379 | # mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \ | |
f6e07d38 | 380 | xxx /sys/fs/cgroup/rg1 |
b6719ec1 LZ |
381 | |
382 | Note that specifying 'release_agent' more than once will return failure. | |
ddbcc7e8 | 383 | |
26d5bbe5 TH |
384 | Note that changing the set of subsystems is currently only supported |
385 | when the hierarchy consists of a single (root) cgroup. Supporting | |
386 | the ability to arbitrarily bind/unbind subsystems from an existing | |
387 | cgroup hierarchy is intended to be implemented in the future. | |
ddbcc7e8 | 388 | |
f6e07d38 JS |
389 | Then under /sys/fs/cgroup/rg1 you can find a tree that corresponds to the |
390 | tree of the cgroups in the system. For instance, /sys/fs/cgroup/rg1 | |
ddbcc7e8 PM |
391 | is the cgroup that holds the whole system. |
392 | ||
b6719ec1 | 393 | If you want to change the value of release_agent: |
f6e07d38 | 394 | # echo "/sbin/new_release_agent" > /sys/fs/cgroup/rg1/release_agent |
b6719ec1 LZ |
395 | |
396 | It can also be changed via remount. | |
397 | ||
f6e07d38 JS |
398 | If you want to create a new cgroup under /sys/fs/cgroup/rg1: |
399 | # cd /sys/fs/cgroup/rg1 | |
ddbcc7e8 PM |
400 | # mkdir my_cgroup |
401 | ||
402 | Now you want to do something with this cgroup. | |
403 | # cd my_cgroup | |
404 | ||
405 | In this directory you can find several files: | |
406 | # ls | |
7823da36 | 407 | cgroup.procs notify_on_release tasks |
d19e0583 | 408 | (plus whatever files added by the attached subsystems) |
ddbcc7e8 PM |
409 | |
410 | Now attach your shell to this cgroup: | |
411 | # /bin/echo $$ > tasks | |
412 | ||
413 | You can also create cgroups inside your cgroup by using mkdir in this | |
414 | directory. | |
415 | # mkdir my_sub_cs | |
416 | ||
417 | To remove a cgroup, just use rmdir: | |
418 | # rmdir my_sub_cs | |
419 | ||
420 | This will fail if the cgroup is in use (has cgroups inside, or | |
421 | has processes attached, or is held alive by other subsystem-specific | |
422 | reference). | |
423 | ||
424 | 2.2 Attaching processes | |
425 | ----------------------- | |
426 | ||
427 | # /bin/echo PID > tasks | |
428 | ||
429 | Note that it is PID, not PIDs. You can only attach ONE task at a time. | |
430 | If you have several tasks to attach, you have to do it one after another: | |
431 | ||
432 | # /bin/echo PID1 > tasks | |
433 | # /bin/echo PID2 > tasks | |
434 | ... | |
435 | # /bin/echo PIDn > tasks | |
436 | ||
bef67c5a LZ |
437 | You can attach the current shell task by echoing 0: |
438 | ||
439 | # echo 0 > tasks | |
440 | ||
74a1166d | 441 | You can use the cgroup.procs file instead of the tasks file to move all |
83b061fc | 442 | threads in a threadgroup at once. Echoing the PID of any task in a |
74a1166d | 443 | threadgroup to cgroup.procs causes all tasks in that threadgroup to be |
1ae65ae9 | 444 | attached to the cgroup. Writing 0 to cgroup.procs moves all tasks |
74a1166d BB |
445 | in the writing task's threadgroup. |
446 | ||
bb6405ea EM |
447 | Note: Since every task is always a member of exactly one cgroup in each |
448 | mounted hierarchy, to remove a task from its current cgroup you must | |
449 | move it into a new cgroup (possibly the root cgroup) by writing to the | |
450 | new cgroup's tasks file. | |
451 | ||
5fe69d7e LZ |
452 | Note: Due to some restrictions enforced by some cgroup subsystems, moving |
453 | a process to another cgroup can fail. | |
bb6405ea | 454 | |
c6d57f33 PM |
455 | 2.3 Mounting hierarchies by name |
456 | -------------------------------- | |
457 | ||
458 | Passing the name=<x> option when mounting a cgroups hierarchy | |
459 | associates the given name with the hierarchy. This can be used when | |
460 | mounting a pre-existing hierarchy, in order to refer to it by name | |
461 | rather than by its set of active subsystems. Each hierarchy is either | |
462 | nameless, or has a unique name. | |
463 | ||
464 | The name should match [\w.-]+ | |
465 | ||
466 | When passing a name=<x> option for a new hierarchy, you need to | |
467 | specify subsystems manually; the legacy behaviour of mounting all | |
468 | subsystems when none are explicitly specified is not supported when | |
469 | you give a subsystem a name. | |
470 | ||
471 | The name of the subsystem appears as part of the hierarchy description | |
472 | in /proc/mounts and /proc/<pid>/cgroups. | |
473 | ||
474 | ||
ddbcc7e8 PM |
475 | 3. Kernel API |
476 | ============= | |
477 | ||
478 | 3.1 Overview | |
479 | ------------ | |
480 | ||
481 | Each kernel subsystem that wants to hook into the generic cgroup | |
482 | system needs to create a cgroup_subsys object. This contains | |
483 | various methods, which are callbacks from the cgroup system, along | |
83b061fc | 484 | with a subsystem ID which will be assigned by the cgroup system. |
ddbcc7e8 PM |
485 | |
486 | Other fields in the cgroup_subsys object include: | |
487 | ||
488 | - subsys_id: a unique array index for the subsystem, indicating which | |
d19e0583 | 489 | entry in cgroup->subsys[] this subsystem should be managing. |
ddbcc7e8 | 490 | |
d19e0583 LZ |
491 | - name: should be initialized to a unique subsystem name. Should be |
492 | no longer than MAX_CGROUP_TYPE_NAMELEN. | |
ddbcc7e8 | 493 | |
d19e0583 LZ |
494 | - early_init: indicate if the subsystem needs early initialization |
495 | at system boot. | |
ddbcc7e8 PM |
496 | |
497 | Each cgroup object created by the system has an array of pointers, | |
83b061fc | 498 | indexed by subsystem ID; this pointer is entirely managed by the |
ddbcc7e8 PM |
499 | subsystem; the generic cgroup code will never touch this pointer. |
500 | ||
501 | 3.2 Synchronization | |
502 | ------------------- | |
503 | ||
504 | There is a global mutex, cgroup_mutex, used by the cgroup | |
505 | system. This should be taken by anything that wants to modify a | |
506 | cgroup. It may also be taken to prevent cgroups from being | |
507 | modified, but more specific locks may be more appropriate in that | |
508 | situation. | |
509 | ||
510 | See kernel/cgroup.c for more details. | |
511 | ||
512 | Subsystems can take/release the cgroup_mutex via the functions | |
ddbcc7e8 PM |
513 | cgroup_lock()/cgroup_unlock(). |
514 | ||
515 | Accessing a task's cgroup pointer may be done in the following ways: | |
516 | - while holding cgroup_mutex | |
517 | - while holding the task's alloc_lock (via task_lock()) | |
518 | - inside an rcu_read_lock() section via rcu_dereference() | |
519 | ||
520 | 3.3 Subsystem API | |
d19e0583 | 521 | ----------------- |
ddbcc7e8 PM |
522 | |
523 | Each subsystem should: | |
524 | ||
525 | - add an entry in linux/cgroup_subsys.h | |
526 | - define a cgroup_subsys object called <name>_subsys | |
527 | ||
e6a1105b | 528 | If a subsystem can be compiled as a module, it should also have in its |
cf5d5941 BB |
529 | module initcall a call to cgroup_load_subsys(), and in its exitcall a |
530 | call to cgroup_unload_subsys(). It should also set its_subsys.module = | |
531 | THIS_MODULE in its .c file. | |
e6a1105b | 532 | |
ddbcc7e8 | 533 | Each subsystem may export the following methods. The only mandatory |
92fb9748 | 534 | methods are css_alloc/free. Any others that are null are presumed to |
ddbcc7e8 PM |
535 | be successful no-ops. |
536 | ||
92fb9748 | 537 | struct cgroup_subsys_state *css_alloc(struct cgroup *cgrp) |
8dc4f3e1 | 538 | (cgroup_mutex held by caller) |
ddbcc7e8 | 539 | |
92fb9748 | 540 | Called to allocate a subsystem state object for a cgroup. The |
ddbcc7e8 PM |
541 | subsystem should allocate its subsystem state object for the passed |
542 | cgroup, returning a pointer to the new object on success or a | |
92fb9748 | 543 | ERR_PTR() value. On success, the subsystem pointer should point to |
ddbcc7e8 PM |
544 | a structure of type cgroup_subsys_state (typically embedded in a |
545 | larger subsystem-specific object), which will be initialized by the | |
546 | cgroup system. Note that this will be called at initialization to | |
547 | create the root subsystem state for this subsystem; this case can be | |
548 | identified by the passed cgroup object having a NULL parent (since | |
549 | it's the root of the hierarchy) and may be an appropriate place for | |
550 | initialization code. | |
551 | ||
92fb9748 | 552 | int css_online(struct cgroup *cgrp) |
8dc4f3e1 | 553 | (cgroup_mutex held by caller) |
ddbcc7e8 | 554 | |
92fb9748 TH |
555 | Called after @cgrp successfully completed all allocations and made |
556 | visible to cgroup_for_each_child/descendant_*() iterators. The | |
557 | subsystem may choose to fail creation by returning -errno. This | |
558 | callback can be used to implement reliable state sharing and | |
559 | propagation along the hierarchy. See the comment on | |
560 | cgroup_for_each_descendant_pre() for details. | |
561 | ||
562 | void css_offline(struct cgroup *cgrp); | |
d7eeac19 | 563 | (cgroup_mutex held by caller) |
92fb9748 TH |
564 | |
565 | This is the counterpart of css_online() and called iff css_online() | |
566 | has succeeded on @cgrp. This signifies the beginning of the end of | |
567 | @cgrp. @cgrp is being removed and the subsystem should start dropping | |
568 | all references it's holding on @cgrp. When all references are dropped, | |
569 | cgroup removal will proceed to the next step - css_free(). After this | |
570 | callback, @cgrp should be considered dead to the subsystem. | |
571 | ||
572 | void css_free(struct cgroup *cgrp) | |
573 | (cgroup_mutex held by caller) | |
574 | ||
575 | The cgroup system is about to free @cgrp; the subsystem should free | |
576 | its subsystem state object. By the time this method is called, @cgrp | |
577 | is completely unused; @cgrp->parent is still valid. (Note - can also | |
578 | be called for a newly-created cgroup if an error occurs after this | |
579 | subsystem's create() method has been called for the new cgroup). | |
d19e0583 | 580 | |
761b3ef5 | 581 | int can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
8dc4f3e1 | 582 | (cgroup_mutex held by caller) |
ddbcc7e8 | 583 | |
2f7ee569 TH |
584 | Called prior to moving one or more tasks into a cgroup; if the |
585 | subsystem returns an error, this will abort the attach operation. | |
586 | @tset contains the tasks to be attached and is guaranteed to have at | |
587 | least one task in it. | |
588 | ||
589 | If there are multiple tasks in the taskset, then: | |
590 | - it's guaranteed that all are from the same thread group | |
591 | - @tset contains all tasks from the thread group whether or not | |
592 | they're switching cgroups | |
593 | - the first task is the leader | |
594 | ||
595 | Each @tset entry also contains the task's old cgroup and tasks which | |
596 | aren't switching cgroup can be skipped easily using the | |
597 | cgroup_taskset_for_each() iterator. Note that this isn't called on a | |
598 | fork. If this method returns 0 (success) then this should remain valid | |
599 | while the caller holds cgroup_mutex and it is ensured that either | |
f780bdb7 BB |
600 | attach() or cancel_attach() will be called in future. |
601 | ||
b4536f0c TH |
602 | void css_reset(struct cgroup_subsys_state *css) |
603 | (cgroup_mutex held by caller) | |
604 | ||
605 | An optional operation which should restore @css's configuration to the | |
606 | initial state. This is currently only used on the unified hierarchy | |
607 | when a subsystem is disabled on a cgroup through | |
608 | "cgroup.subtree_control" but should remain enabled because other | |
609 | subsystems depend on it. cgroup core makes such a css invisible by | |
610 | removing the associated interface files and invokes this callback so | |
611 | that the hidden subsystem can return to the initial neutral state. | |
612 | This prevents unexpected resource control from a hidden css and | |
613 | ensures that the configuration is in the initial state when it is made | |
614 | visible again later. | |
615 | ||
761b3ef5 | 616 | void cancel_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
2468c723 DN |
617 | (cgroup_mutex held by caller) |
618 | ||
619 | Called when a task attach operation has failed after can_attach() has succeeded. | |
620 | A subsystem whose can_attach() has some side-effects should provide this | |
88393161 | 621 | function, so that the subsystem can implement a rollback. If not, not necessary. |
2468c723 | 622 | This will be called only about subsystems whose can_attach() operation have |
2f7ee569 | 623 | succeeded. The parameters are identical to can_attach(). |
2468c723 | 624 | |
761b3ef5 | 625 | void attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
18e7f1f0 | 626 | (cgroup_mutex held by caller) |
ddbcc7e8 PM |
627 | |
628 | Called after the task has been attached to the cgroup, to allow any | |
629 | post-attachment activity that requires memory allocations or blocking. | |
2f7ee569 | 630 | The parameters are identical to can_attach(). |
f780bdb7 | 631 | |
761b3ef5 | 632 | void fork(struct task_struct *task) |
ddbcc7e8 | 633 | |
e8d55fde | 634 | Called when a task is forked into a cgroup. |
ddbcc7e8 | 635 | |
761b3ef5 | 636 | void exit(struct task_struct *task) |
ddbcc7e8 | 637 | |
d19e0583 | 638 | Called during task exit. |
ddbcc7e8 | 639 | |
afcf6c8b TH |
640 | void free(struct task_struct *task) |
641 | ||
642 | Called when the task_struct is freed. | |
643 | ||
26d5bbe5 TH |
644 | void bind(struct cgroup *root) |
645 | (cgroup_mutex held by caller) | |
646 | ||
647 | Called when a cgroup subsystem is rebound to a different hierarchy | |
648 | and root cgroup. Currently this will only involve movement between | |
649 | the default hierarchy (which never has sub-cgroups) and a hierarchy | |
650 | that is being created/destroyed (and hence has no sub-cgroups). | |
ddbcc7e8 | 651 | |
19ec2567 AR |
652 | 4. Extended attribute usage |
653 | =========================== | |
654 | ||
655 | cgroup filesystem supports certain types of extended attributes in its | |
656 | directories and files. The current supported types are: | |
657 | - Trusted (XATTR_TRUSTED) | |
658 | - Security (XATTR_SECURITY) | |
659 | ||
660 | Both require CAP_SYS_ADMIN capability to set. | |
661 | ||
662 | Like in tmpfs, the extended attributes in cgroup filesystem are stored | |
663 | using kernel memory and it's advised to keep the usage at minimum. This | |
664 | is the reason why user defined extended attributes are not supported, since | |
665 | any user can do it and there's no limit in the value size. | |
666 | ||
667 | The current known users for this feature are SELinux to limit cgroup usage | |
668 | in containers and systemd for assorted meta data like main PID in a cgroup | |
669 | (systemd creates a cgroup per service). | |
670 | ||
671 | 5. Questions | |
ddbcc7e8 PM |
672 | ============ |
673 | ||
674 | Q: what's up with this '/bin/echo' ? | |
675 | A: bash's builtin 'echo' command does not check calls to write() against | |
676 | errors. If you use it in the cgroup file system, you won't be | |
677 | able to tell whether a command succeeded or failed. | |
678 | ||
679 | Q: When I attach processes, only the first of the line gets really attached ! | |
680 | A: We can only return one error code per call to write(). So you should also | |
83b061fc | 681 | put only ONE PID. |
ddbcc7e8 | 682 |