]>
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> | |
11 | Modified by Christoph Lameter <clameter@sgi.com> | |
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 |
0dea1168 | 27 | 2.4 Notification API |
ddbcc7e8 PM |
28 | 3. Kernel API |
29 | 3.1 Overview | |
30 | 3.2 Synchronization | |
31 | 3.3 Subsystem API | |
19ec2567 AR |
32 | 4. Extended attributes usage |
33 | 5. Questions | |
ddbcc7e8 PM |
34 | |
35 | 1. Control Groups | |
d19e0583 | 36 | ================= |
ddbcc7e8 PM |
37 | |
38 | 1.1 What are cgroups ? | |
39 | ---------------------- | |
40 | ||
41 | Control Groups provide a mechanism for aggregating/partitioning sets of | |
42 | tasks, and all their future children, into hierarchical groups with | |
43 | specialized behaviour. | |
44 | ||
45 | Definitions: | |
46 | ||
47 | A *cgroup* associates a set of tasks with a set of parameters for one | |
48 | or more subsystems. | |
49 | ||
50 | A *subsystem* is a module that makes use of the task grouping | |
51 | facilities provided by cgroups to treat groups of tasks in | |
52 | particular ways. A subsystem is typically a "resource controller" that | |
53 | schedules a resource or applies per-cgroup limits, but it may be | |
54 | anything that wants to act on a group of processes, e.g. a | |
55 | virtualization subsystem. | |
56 | ||
57 | A *hierarchy* is a set of cgroups arranged in a tree, such that | |
58 | every task in the system is in exactly one of the cgroups in the | |
59 | hierarchy, and a set of subsystems; each subsystem has system-specific | |
60 | state attached to each cgroup in the hierarchy. Each hierarchy has | |
61 | an instance of the cgroup virtual filesystem associated with it. | |
62 | ||
caa790ba | 63 | At any one time there may be multiple active hierarchies of task |
ddbcc7e8 PM |
64 | cgroups. Each hierarchy is a partition of all tasks in the system. |
65 | ||
83b061fc | 66 | User-level code may create and destroy cgroups by name in an |
ddbcc7e8 | 67 | instance of the cgroup virtual file system, specify and query to |
83b061fc | 68 | which cgroup a task is assigned, and list the task PIDs assigned to |
ddbcc7e8 PM |
69 | a cgroup. Those creations and assignments only affect the hierarchy |
70 | associated with that instance of the cgroup file system. | |
71 | ||
72 | On their own, the only use for cgroups is for simple job | |
73 | tracking. The intention is that other subsystems hook into the generic | |
74 | cgroup support to provide new attributes for cgroups, such as | |
75 | accounting/limiting the resources which processes in a cgroup can | |
83b061fc | 76 | access. For example, cpusets (see Documentation/cgroups/cpusets.txt) allow |
ddbcc7e8 PM |
77 | you to associate a set of CPUs and a set of memory nodes with the |
78 | tasks in each cgroup. | |
79 | ||
80 | 1.2 Why are cgroups needed ? | |
81 | ---------------------------- | |
82 | ||
83 | There are multiple efforts to provide process aggregations in the | |
83b061fc | 84 | Linux kernel, mainly for resource-tracking purposes. Such efforts |
ddbcc7e8 PM |
85 | include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server |
86 | namespaces. These all require the basic notion of a | |
87 | grouping/partitioning of processes, with newly forked processes ending | |
83b061fc | 88 | up in the same group (cgroup) as their parent process. |
ddbcc7e8 PM |
89 | |
90 | The kernel cgroup patch provides the minimum essential kernel | |
91 | mechanisms required to efficiently implement such groups. It has | |
92 | minimal impact on the system fast paths, and provides hooks for | |
93 | specific subsystems such as cpusets to provide additional behaviour as | |
94 | desired. | |
95 | ||
96 | Multiple hierarchy support is provided to allow for situations where | |
97 | the division of tasks into cgroups is distinctly different for | |
98 | different subsystems - having parallel hierarchies allows each | |
99 | hierarchy to be a natural division of tasks, without having to handle | |
100 | complex combinations of tasks that would be present if several | |
101 | unrelated subsystems needed to be forced into the same tree of | |
102 | cgroups. | |
103 | ||
104 | At one extreme, each resource controller or subsystem could be in a | |
105 | separate hierarchy; at the other extreme, all subsystems | |
106 | would be attached to the same hierarchy. | |
107 | ||
108 | As an example of a scenario (originally proposed by vatsa@in.ibm.com) | |
109 | that can benefit from multiple hierarchies, consider a large | |
110 | university server with various users - students, professors, system | |
111 | tasks etc. The resource planning for this server could be along the | |
112 | following lines: | |
113 | ||
6ad85239 | 114 | CPU : "Top cpuset" |
ddbcc7e8 PM |
115 | / \ |
116 | CPUSet1 CPUSet2 | |
6ad85239 GL |
117 | | | |
118 | (Professors) (Students) | |
ddbcc7e8 PM |
119 | |
120 | In addition (system tasks) are attached to topcpuset (so | |
121 | that they can run anywhere) with a limit of 20% | |
122 | ||
6ad85239 | 123 | Memory : Professors (50%), Students (30%), system (20%) |
ddbcc7e8 | 124 | |
6ad85239 | 125 | Disk : Professors (50%), Students (30%), system (20%) |
ddbcc7e8 PM |
126 | |
127 | Network : WWW browsing (20%), Network File System (60%), others (20%) | |
128 | / \ | |
6ad85239 | 129 | Professors (15%) students (5%) |
ddbcc7e8 | 130 | |
83b061fc MK |
131 | Browsers like Firefox/Lynx go into the WWW network class, while (k)nfsd goes |
132 | into the NFS network class. | |
ddbcc7e8 | 133 | |
caa790ba | 134 | At the same time Firefox/Lynx will share an appropriate CPU/Memory class |
ddbcc7e8 PM |
135 | depending on who launched it (prof/student). |
136 | ||
137 | With the ability to classify tasks differently for different resources | |
83b061fc | 138 | (by putting those resource subsystems in different hierarchies), |
ddbcc7e8 PM |
139 | the admin can easily set up a script which receives exec notifications |
140 | and depending on who is launching the browser he can | |
141 | ||
f6e07d38 | 142 | # echo browser_pid > /sys/fs/cgroup/<restype>/<userclass>/tasks |
ddbcc7e8 PM |
143 | |
144 | With only a single hierarchy, he now would potentially have to create | |
145 | a separate cgroup for every browser launched and associate it with | |
67de0162 | 146 | appropriate network and other resource class. This may lead to |
ddbcc7e8 PM |
147 | proliferation of such cgroups. |
148 | ||
83b061fc | 149 | Also let's say that the administrator would like to give enhanced network |
ddbcc7e8 | 150 | access temporarily to a student's browser (since it is night and the user |
83b061fc MK |
151 | wants to do online gaming :)) OR give one of the student's simulation |
152 | apps enhanced CPU power. | |
ddbcc7e8 | 153 | |
83b061fc MK |
154 | With ability to write PIDs directly to resource classes, it's just a |
155 | matter of: | |
ddbcc7e8 | 156 | |
f6e07d38 | 157 | # echo pid > /sys/fs/cgroup/network/<new_class>/tasks |
ddbcc7e8 | 158 | (after some time) |
f6e07d38 | 159 | # echo pid > /sys/fs/cgroup/network/<orig_class>/tasks |
ddbcc7e8 | 160 | |
83b061fc | 161 | Without this ability, the administrator would have to split the cgroup into |
ddbcc7e8 PM |
162 | multiple separate ones and then associate the new cgroups with the |
163 | new resource classes. | |
164 | ||
165 | ||
166 | ||
167 | 1.3 How are cgroups implemented ? | |
168 | --------------------------------- | |
169 | ||
170 | Control Groups extends the kernel as follows: | |
171 | ||
172 | - Each task in the system has a reference-counted pointer to a | |
173 | css_set. | |
174 | ||
175 | - A css_set contains a set of reference-counted pointers to | |
176 | cgroup_subsys_state objects, one for each cgroup subsystem | |
177 | registered in the system. There is no direct link from a task to | |
178 | the cgroup of which it's a member in each hierarchy, but this | |
179 | can be determined by following pointers through the | |
180 | cgroup_subsys_state objects. This is because accessing the | |
181 | subsystem state is something that's expected to happen frequently | |
182 | and in performance-critical code, whereas operations that require a | |
183 | task's actual cgroup assignments (in particular, moving between | |
817929ec PM |
184 | cgroups) are less common. A linked list runs through the cg_list |
185 | field of each task_struct using the css_set, anchored at | |
186 | css_set->tasks. | |
ddbcc7e8 | 187 | |
83b061fc | 188 | - A cgroup hierarchy filesystem can be mounted for browsing and |
ddbcc7e8 PM |
189 | manipulation from user space. |
190 | ||
83b061fc | 191 | - You can list all the tasks (by PID) attached to any cgroup. |
ddbcc7e8 PM |
192 | |
193 | The implementation of cgroups requires a few, simple hooks | |
83b061fc | 194 | into the rest of the kernel, none in performance-critical paths: |
ddbcc7e8 PM |
195 | |
196 | - in init/main.c, to initialize the root cgroups and initial | |
197 | css_set at system boot. | |
198 | ||
199 | - in fork and exit, to attach and detach a task from its css_set. | |
200 | ||
83b061fc | 201 | In addition, a new file system of type "cgroup" may be mounted, to |
ddbcc7e8 PM |
202 | enable browsing and modifying the cgroups presently known to the |
203 | kernel. When mounting a cgroup hierarchy, you may specify a | |
204 | comma-separated list of subsystems to mount as the filesystem mount | |
205 | options. By default, mounting the cgroup filesystem attempts to | |
206 | mount a hierarchy containing all registered subsystems. | |
207 | ||
208 | If an active hierarchy with exactly the same set of subsystems already | |
209 | exists, it will be reused for the new mount. If no existing hierarchy | |
210 | matches, and any of the requested subsystems are in use in an existing | |
211 | hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy | |
212 | is activated, associated with the requested subsystems. | |
213 | ||
26d5bbe5 TH |
214 | It's not currently possible to bind a new subsystem to an active |
215 | cgroup hierarchy, or to unbind a subsystem from an active cgroup | |
216 | hierarchy. This may be possible in future, but is fraught with nasty | |
217 | error-recovery issues. | |
ddbcc7e8 PM |
218 | |
219 | When a cgroup filesystem is unmounted, if there are any | |
220 | child cgroups created below the top-level cgroup, that hierarchy | |
221 | will remain active even though unmounted; if there are no | |
222 | child cgroups then the hierarchy will be deactivated. | |
223 | ||
224 | No new system calls are added for cgroups - all support for | |
225 | querying and modifying cgroups is via this cgroup file system. | |
226 | ||
227 | Each task under /proc has an added file named 'cgroup' displaying, | |
228 | for each active hierarchy, the subsystem names and the cgroup name | |
229 | as the path relative to the root of the cgroup file system. | |
230 | ||
231 | Each cgroup is represented by a directory in the cgroup file system | |
232 | containing the following files describing that cgroup: | |
233 | ||
83b061fc MK |
234 | - tasks: list of tasks (by PID) attached to that cgroup. This list |
235 | is not guaranteed to be sorted. Writing a thread ID into this file | |
7823da36 | 236 | moves the thread into this cgroup. |
83b061fc MK |
237 | - cgroup.procs: list of thread group IDs in the cgroup. This list is |
238 | not guaranteed to be sorted or free of duplicate TGIDs, and userspace | |
7823da36 | 239 | should sort/uniquify the list if this property is required. |
83b061fc | 240 | Writing a thread group ID into this file moves all threads in that |
74a1166d | 241 | group into this cgroup. |
d19e0583 LZ |
242 | - notify_on_release flag: run the release agent on exit? |
243 | - release_agent: the path to use for release notifications (this file | |
244 | exists in the top cgroup only) | |
ddbcc7e8 PM |
245 | |
246 | Other subsystems such as cpusets may add additional files in each | |
d19e0583 | 247 | cgroup dir. |
ddbcc7e8 PM |
248 | |
249 | New cgroups are created using the mkdir system call or shell | |
250 | command. The properties of a cgroup, such as its flags, are | |
251 | modified by writing to the appropriate file in that cgroups | |
252 | directory, as listed above. | |
253 | ||
254 | The named hierarchical structure of nested cgroups allows partitioning | |
255 | a large system into nested, dynamically changeable, "soft-partitions". | |
256 | ||
257 | The attachment of each task, automatically inherited at fork by any | |
258 | children of that task, to a cgroup allows organizing the work load | |
259 | on a system into related sets of tasks. A task may be re-attached to | |
260 | any other cgroup, if allowed by the permissions on the necessary | |
261 | cgroup file system directories. | |
262 | ||
263 | When a task is moved from one cgroup to another, it gets a new | |
264 | css_set pointer - if there's an already existing css_set with the | |
83b061fc | 265 | desired collection of cgroups then that group is reused, otherwise a new |
b851ee79 LZ |
266 | css_set is allocated. The appropriate existing css_set is located by |
267 | looking into a hash table. | |
ddbcc7e8 | 268 | |
817929ec PM |
269 | To allow access from a cgroup to the css_sets (and hence tasks) |
270 | that comprise it, a set of cg_cgroup_link objects form a lattice; | |
271 | each cg_cgroup_link is linked into a list of cg_cgroup_links for | |
d19e0583 | 272 | a single cgroup on its cgrp_link_list field, and a list of |
817929ec PM |
273 | cg_cgroup_links for a single css_set on its cg_link_list. |
274 | ||
275 | Thus the set of tasks in a cgroup can be listed by iterating over | |
276 | each css_set that references the cgroup, and sub-iterating over | |
277 | each css_set's task set. | |
278 | ||
ddbcc7e8 PM |
279 | The use of a Linux virtual file system (vfs) to represent the |
280 | cgroup hierarchy provides for a familiar permission and name space | |
281 | for cgroups, with a minimum of additional kernel code. | |
282 | ||
283 | 1.4 What does notify_on_release do ? | |
284 | ------------------------------------ | |
285 | ||
ddbcc7e8 PM |
286 | If the notify_on_release flag is enabled (1) in a cgroup, then |
287 | whenever the last task in the cgroup leaves (exits or attaches to | |
288 | some other cgroup) and the last child cgroup of that cgroup | |
289 | is removed, then the kernel runs the command specified by the contents | |
290 | of the "release_agent" file in that hierarchy's root directory, | |
291 | supplying the pathname (relative to the mount point of the cgroup | |
292 | file system) of the abandoned cgroup. This enables automatic | |
293 | removal of abandoned cgroups. The default value of | |
294 | notify_on_release in the root cgroup at system boot is disabled | |
295 | (0). The default value of other cgroups at creation is the current | |
83b061fc | 296 | value of their parents' notify_on_release settings. The default value of |
ddbcc7e8 PM |
297 | a cgroup hierarchy's release_agent path is empty. |
298 | ||
97978e6d DL |
299 | 1.5 What does clone_children do ? |
300 | --------------------------------- | |
301 | ||
2260e7fc TH |
302 | This flag only affects the cpuset controller. If the clone_children |
303 | flag is enabled (1) in a cgroup, a new cpuset cgroup will copy its | |
304 | configuration from the parent during initialization. | |
97978e6d DL |
305 | |
306 | 1.6 How do I use cgroups ? | |
ddbcc7e8 PM |
307 | -------------------------- |
308 | ||
309 | To start a new job that is to be contained within a cgroup, using | |
310 | the "cpuset" cgroup subsystem, the steps are something like: | |
311 | ||
f6e07d38 JS |
312 | 1) mount -t tmpfs cgroup_root /sys/fs/cgroup |
313 | 2) mkdir /sys/fs/cgroup/cpuset | |
314 | 3) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset | |
315 | 4) Create the new cgroup by doing mkdir's and write's (or echo's) in | |
316 | the /sys/fs/cgroup virtual file system. | |
317 | 5) Start a task that will be the "founding father" of the new job. | |
83b061fc | 318 | 6) Attach that task to the new cgroup by writing its PID to the |
f6e07d38 JS |
319 | /sys/fs/cgroup/cpuset/tasks file for that cgroup. |
320 | 7) fork, exec or clone the job tasks from this founding father task. | |
ddbcc7e8 PM |
321 | |
322 | For example, the following sequence of commands will setup a cgroup | |
323 | named "Charlie", containing just CPUs 2 and 3, and Memory Node 1, | |
324 | and then start a subshell 'sh' in that cgroup: | |
325 | ||
f6e07d38 JS |
326 | mount -t tmpfs cgroup_root /sys/fs/cgroup |
327 | mkdir /sys/fs/cgroup/cpuset | |
328 | mount -t cgroup cpuset -ocpuset /sys/fs/cgroup/cpuset | |
329 | cd /sys/fs/cgroup/cpuset | |
ddbcc7e8 PM |
330 | mkdir Charlie |
331 | cd Charlie | |
0f146a76 DG |
332 | /bin/echo 2-3 > cpuset.cpus |
333 | /bin/echo 1 > cpuset.mems | |
ddbcc7e8 PM |
334 | /bin/echo $$ > tasks |
335 | sh | |
336 | # The subshell 'sh' is now running in cgroup Charlie | |
337 | # The next line should display '/Charlie' | |
338 | cat /proc/self/cgroup | |
339 | ||
340 | 2. Usage Examples and Syntax | |
341 | ============================ | |
342 | ||
343 | 2.1 Basic Usage | |
344 | --------------- | |
345 | ||
83b061fc | 346 | Creating, modifying, using cgroups can be done through the cgroup |
ddbcc7e8 PM |
347 | virtual filesystem. |
348 | ||
caa790ba | 349 | To mount a cgroup hierarchy with all available subsystems, type: |
f6e07d38 | 350 | # mount -t cgroup xxx /sys/fs/cgroup |
ddbcc7e8 PM |
351 | |
352 | The "xxx" is not interpreted by the cgroup code, but will appear in | |
353 | /proc/mounts so may be any useful identifying string that you like. | |
354 | ||
bb6405ea EM |
355 | Note: Some subsystems do not work without some user input first. For instance, |
356 | if cpusets are enabled the user will have to populate the cpus and mems files | |
357 | for each new cgroup created before that group can be used. | |
358 | ||
f6e07d38 JS |
359 | As explained in section `1.2 Why are cgroups needed?' you should create |
360 | different hierarchies of cgroups for each single resource or group of | |
361 | resources you want to control. Therefore, you should mount a tmpfs on | |
362 | /sys/fs/cgroup and create directories for each cgroup resource or resource | |
363 | group. | |
364 | ||
365 | # mount -t tmpfs cgroup_root /sys/fs/cgroup | |
366 | # mkdir /sys/fs/cgroup/rg1 | |
367 | ||
595f4b69 | 368 | To mount a cgroup hierarchy with just the cpuset and memory |
ddbcc7e8 | 369 | subsystems, type: |
f6e07d38 | 370 | # mount -t cgroup -o cpuset,memory hier1 /sys/fs/cgroup/rg1 |
ddbcc7e8 | 371 | |
9a8054aa DW |
372 | While remounting cgroups is currently supported, it is not recommend |
373 | to use it. Remounting allows changing bound subsystems and | |
374 | release_agent. Rebinding is hardly useful as it only works when the | |
375 | hierarchy is empty and release_agent itself should be replaced with | |
376 | conventional fsnotify. The support for remounting will be removed in | |
377 | the future. | |
b6719ec1 LZ |
378 | |
379 | To Specify a hierarchy's release_agent: | |
380 | # mount -t cgroup -o cpuset,release_agent="/sbin/cpuset_release_agent" \ | |
f6e07d38 | 381 | xxx /sys/fs/cgroup/rg1 |
b6719ec1 LZ |
382 | |
383 | Note that specifying 'release_agent' more than once will return failure. | |
ddbcc7e8 | 384 | |
26d5bbe5 TH |
385 | Note that changing the set of subsystems is currently only supported |
386 | when the hierarchy consists of a single (root) cgroup. Supporting | |
387 | the ability to arbitrarily bind/unbind subsystems from an existing | |
388 | cgroup hierarchy is intended to be implemented in the future. | |
ddbcc7e8 | 389 | |
f6e07d38 JS |
390 | Then under /sys/fs/cgroup/rg1 you can find a tree that corresponds to the |
391 | tree of the cgroups in the system. For instance, /sys/fs/cgroup/rg1 | |
ddbcc7e8 PM |
392 | is the cgroup that holds the whole system. |
393 | ||
b6719ec1 | 394 | If you want to change the value of release_agent: |
f6e07d38 | 395 | # echo "/sbin/new_release_agent" > /sys/fs/cgroup/rg1/release_agent |
b6719ec1 LZ |
396 | |
397 | It can also be changed via remount. | |
398 | ||
f6e07d38 JS |
399 | If you want to create a new cgroup under /sys/fs/cgroup/rg1: |
400 | # cd /sys/fs/cgroup/rg1 | |
ddbcc7e8 PM |
401 | # mkdir my_cgroup |
402 | ||
403 | Now you want to do something with this cgroup. | |
404 | # cd my_cgroup | |
405 | ||
406 | In this directory you can find several files: | |
407 | # ls | |
7823da36 | 408 | cgroup.procs notify_on_release tasks |
d19e0583 | 409 | (plus whatever files added by the attached subsystems) |
ddbcc7e8 PM |
410 | |
411 | Now attach your shell to this cgroup: | |
412 | # /bin/echo $$ > tasks | |
413 | ||
414 | You can also create cgroups inside your cgroup by using mkdir in this | |
415 | directory. | |
416 | # mkdir my_sub_cs | |
417 | ||
418 | To remove a cgroup, just use rmdir: | |
419 | # rmdir my_sub_cs | |
420 | ||
421 | This will fail if the cgroup is in use (has cgroups inside, or | |
422 | has processes attached, or is held alive by other subsystem-specific | |
423 | reference). | |
424 | ||
425 | 2.2 Attaching processes | |
426 | ----------------------- | |
427 | ||
428 | # /bin/echo PID > tasks | |
429 | ||
430 | Note that it is PID, not PIDs. You can only attach ONE task at a time. | |
431 | If you have several tasks to attach, you have to do it one after another: | |
432 | ||
433 | # /bin/echo PID1 > tasks | |
434 | # /bin/echo PID2 > tasks | |
435 | ... | |
436 | # /bin/echo PIDn > tasks | |
437 | ||
bef67c5a LZ |
438 | You can attach the current shell task by echoing 0: |
439 | ||
440 | # echo 0 > tasks | |
441 | ||
74a1166d | 442 | You can use the cgroup.procs file instead of the tasks file to move all |
83b061fc | 443 | threads in a threadgroup at once. Echoing the PID of any task in a |
74a1166d | 444 | threadgroup to cgroup.procs causes all tasks in that threadgroup to be |
1ae65ae9 | 445 | attached to the cgroup. Writing 0 to cgroup.procs moves all tasks |
74a1166d BB |
446 | in the writing task's threadgroup. |
447 | ||
bb6405ea EM |
448 | Note: Since every task is always a member of exactly one cgroup in each |
449 | mounted hierarchy, to remove a task from its current cgroup you must | |
450 | move it into a new cgroup (possibly the root cgroup) by writing to the | |
451 | new cgroup's tasks file. | |
452 | ||
5fe69d7e LZ |
453 | Note: Due to some restrictions enforced by some cgroup subsystems, moving |
454 | a process to another cgroup can fail. | |
bb6405ea | 455 | |
c6d57f33 PM |
456 | 2.3 Mounting hierarchies by name |
457 | -------------------------------- | |
458 | ||
459 | Passing the name=<x> option when mounting a cgroups hierarchy | |
460 | associates the given name with the hierarchy. This can be used when | |
461 | mounting a pre-existing hierarchy, in order to refer to it by name | |
462 | rather than by its set of active subsystems. Each hierarchy is either | |
463 | nameless, or has a unique name. | |
464 | ||
465 | The name should match [\w.-]+ | |
466 | ||
467 | When passing a name=<x> option for a new hierarchy, you need to | |
468 | specify subsystems manually; the legacy behaviour of mounting all | |
469 | subsystems when none are explicitly specified is not supported when | |
470 | you give a subsystem a name. | |
471 | ||
472 | The name of the subsystem appears as part of the hierarchy description | |
473 | in /proc/mounts and /proc/<pid>/cgroups. | |
474 | ||
0dea1168 KS |
475 | 2.4 Notification API |
476 | -------------------- | |
477 | ||
478 | There is mechanism which allows to get notifications about changing | |
479 | status of a cgroup. | |
480 | ||
83b061fc | 481 | To register a new notification handler you need to: |
0dea1168 KS |
482 | - create a file descriptor for event notification using eventfd(2); |
483 | - open a control file to be monitored (e.g. memory.usage_in_bytes); | |
484 | - write "<event_fd> <control_fd> <args>" to cgroup.event_control. | |
485 | Interpretation of args is defined by control file implementation; | |
486 | ||
487 | eventfd will be woken up by control file implementation or when the | |
488 | cgroup is removed. | |
489 | ||
83b061fc | 490 | To unregister a notification handler just close eventfd. |
0dea1168 KS |
491 | |
492 | NOTE: Support of notifications should be implemented for the control | |
493 | file. See documentation for the subsystem. | |
c6d57f33 | 494 | |
ddbcc7e8 PM |
495 | 3. Kernel API |
496 | ============= | |
497 | ||
498 | 3.1 Overview | |
499 | ------------ | |
500 | ||
501 | Each kernel subsystem that wants to hook into the generic cgroup | |
502 | system needs to create a cgroup_subsys object. This contains | |
503 | various methods, which are callbacks from the cgroup system, along | |
83b061fc | 504 | with a subsystem ID which will be assigned by the cgroup system. |
ddbcc7e8 PM |
505 | |
506 | Other fields in the cgroup_subsys object include: | |
507 | ||
508 | - subsys_id: a unique array index for the subsystem, indicating which | |
d19e0583 | 509 | entry in cgroup->subsys[] this subsystem should be managing. |
ddbcc7e8 | 510 | |
d19e0583 LZ |
511 | - name: should be initialized to a unique subsystem name. Should be |
512 | no longer than MAX_CGROUP_TYPE_NAMELEN. | |
ddbcc7e8 | 513 | |
d19e0583 LZ |
514 | - early_init: indicate if the subsystem needs early initialization |
515 | at system boot. | |
ddbcc7e8 PM |
516 | |
517 | Each cgroup object created by the system has an array of pointers, | |
83b061fc | 518 | indexed by subsystem ID; this pointer is entirely managed by the |
ddbcc7e8 PM |
519 | subsystem; the generic cgroup code will never touch this pointer. |
520 | ||
521 | 3.2 Synchronization | |
522 | ------------------- | |
523 | ||
524 | There is a global mutex, cgroup_mutex, used by the cgroup | |
525 | system. This should be taken by anything that wants to modify a | |
526 | cgroup. It may also be taken to prevent cgroups from being | |
527 | modified, but more specific locks may be more appropriate in that | |
528 | situation. | |
529 | ||
530 | See kernel/cgroup.c for more details. | |
531 | ||
532 | Subsystems can take/release the cgroup_mutex via the functions | |
ddbcc7e8 PM |
533 | cgroup_lock()/cgroup_unlock(). |
534 | ||
535 | Accessing a task's cgroup pointer may be done in the following ways: | |
536 | - while holding cgroup_mutex | |
537 | - while holding the task's alloc_lock (via task_lock()) | |
538 | - inside an rcu_read_lock() section via rcu_dereference() | |
539 | ||
540 | 3.3 Subsystem API | |
d19e0583 | 541 | ----------------- |
ddbcc7e8 PM |
542 | |
543 | Each subsystem should: | |
544 | ||
545 | - add an entry in linux/cgroup_subsys.h | |
546 | - define a cgroup_subsys object called <name>_subsys | |
547 | ||
e6a1105b | 548 | If a subsystem can be compiled as a module, it should also have in its |
cf5d5941 BB |
549 | module initcall a call to cgroup_load_subsys(), and in its exitcall a |
550 | call to cgroup_unload_subsys(). It should also set its_subsys.module = | |
551 | THIS_MODULE in its .c file. | |
e6a1105b | 552 | |
ddbcc7e8 | 553 | Each subsystem may export the following methods. The only mandatory |
92fb9748 | 554 | methods are css_alloc/free. Any others that are null are presumed to |
ddbcc7e8 PM |
555 | be successful no-ops. |
556 | ||
92fb9748 | 557 | struct cgroup_subsys_state *css_alloc(struct cgroup *cgrp) |
8dc4f3e1 | 558 | (cgroup_mutex held by caller) |
ddbcc7e8 | 559 | |
92fb9748 | 560 | Called to allocate a subsystem state object for a cgroup. The |
ddbcc7e8 PM |
561 | subsystem should allocate its subsystem state object for the passed |
562 | cgroup, returning a pointer to the new object on success or a | |
92fb9748 | 563 | ERR_PTR() value. On success, the subsystem pointer should point to |
ddbcc7e8 PM |
564 | a structure of type cgroup_subsys_state (typically embedded in a |
565 | larger subsystem-specific object), which will be initialized by the | |
566 | cgroup system. Note that this will be called at initialization to | |
567 | create the root subsystem state for this subsystem; this case can be | |
568 | identified by the passed cgroup object having a NULL parent (since | |
569 | it's the root of the hierarchy) and may be an appropriate place for | |
570 | initialization code. | |
571 | ||
92fb9748 | 572 | int css_online(struct cgroup *cgrp) |
8dc4f3e1 | 573 | (cgroup_mutex held by caller) |
ddbcc7e8 | 574 | |
92fb9748 TH |
575 | Called after @cgrp successfully completed all allocations and made |
576 | visible to cgroup_for_each_child/descendant_*() iterators. The | |
577 | subsystem may choose to fail creation by returning -errno. This | |
578 | callback can be used to implement reliable state sharing and | |
579 | propagation along the hierarchy. See the comment on | |
580 | cgroup_for_each_descendant_pre() for details. | |
581 | ||
582 | void css_offline(struct cgroup *cgrp); | |
d7eeac19 | 583 | (cgroup_mutex held by caller) |
92fb9748 TH |
584 | |
585 | This is the counterpart of css_online() and called iff css_online() | |
586 | has succeeded on @cgrp. This signifies the beginning of the end of | |
587 | @cgrp. @cgrp is being removed and the subsystem should start dropping | |
588 | all references it's holding on @cgrp. When all references are dropped, | |
589 | cgroup removal will proceed to the next step - css_free(). After this | |
590 | callback, @cgrp should be considered dead to the subsystem. | |
591 | ||
592 | void css_free(struct cgroup *cgrp) | |
593 | (cgroup_mutex held by caller) | |
594 | ||
595 | The cgroup system is about to free @cgrp; the subsystem should free | |
596 | its subsystem state object. By the time this method is called, @cgrp | |
597 | is completely unused; @cgrp->parent is still valid. (Note - can also | |
598 | be called for a newly-created cgroup if an error occurs after this | |
599 | subsystem's create() method has been called for the new cgroup). | |
d19e0583 | 600 | |
761b3ef5 | 601 | int can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
8dc4f3e1 | 602 | (cgroup_mutex held by caller) |
ddbcc7e8 | 603 | |
2f7ee569 TH |
604 | Called prior to moving one or more tasks into a cgroup; if the |
605 | subsystem returns an error, this will abort the attach operation. | |
606 | @tset contains the tasks to be attached and is guaranteed to have at | |
607 | least one task in it. | |
608 | ||
609 | If there are multiple tasks in the taskset, then: | |
610 | - it's guaranteed that all are from the same thread group | |
611 | - @tset contains all tasks from the thread group whether or not | |
612 | they're switching cgroups | |
613 | - the first task is the leader | |
614 | ||
615 | Each @tset entry also contains the task's old cgroup and tasks which | |
616 | aren't switching cgroup can be skipped easily using the | |
617 | cgroup_taskset_for_each() iterator. Note that this isn't called on a | |
618 | fork. If this method returns 0 (success) then this should remain valid | |
619 | while the caller holds cgroup_mutex and it is ensured that either | |
f780bdb7 BB |
620 | attach() or cancel_attach() will be called in future. |
621 | ||
761b3ef5 | 622 | void cancel_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
2468c723 DN |
623 | (cgroup_mutex held by caller) |
624 | ||
625 | Called when a task attach operation has failed after can_attach() has succeeded. | |
626 | A subsystem whose can_attach() has some side-effects should provide this | |
88393161 | 627 | function, so that the subsystem can implement a rollback. If not, not necessary. |
2468c723 | 628 | This will be called only about subsystems whose can_attach() operation have |
2f7ee569 | 629 | succeeded. The parameters are identical to can_attach(). |
2468c723 | 630 | |
761b3ef5 | 631 | void attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
18e7f1f0 | 632 | (cgroup_mutex held by caller) |
ddbcc7e8 PM |
633 | |
634 | Called after the task has been attached to the cgroup, to allow any | |
635 | post-attachment activity that requires memory allocations or blocking. | |
2f7ee569 | 636 | The parameters are identical to can_attach(). |
f780bdb7 | 637 | |
761b3ef5 | 638 | void fork(struct task_struct *task) |
ddbcc7e8 | 639 | |
e8d55fde | 640 | Called when a task is forked into a cgroup. |
ddbcc7e8 | 641 | |
761b3ef5 | 642 | void exit(struct task_struct *task) |
ddbcc7e8 | 643 | |
d19e0583 | 644 | Called during task exit. |
ddbcc7e8 | 645 | |
26d5bbe5 TH |
646 | void bind(struct cgroup *root) |
647 | (cgroup_mutex held by caller) | |
648 | ||
649 | Called when a cgroup subsystem is rebound to a different hierarchy | |
650 | and root cgroup. Currently this will only involve movement between | |
651 | the default hierarchy (which never has sub-cgroups) and a hierarchy | |
652 | that is being created/destroyed (and hence has no sub-cgroups). | |
ddbcc7e8 | 653 | |
19ec2567 AR |
654 | 4. Extended attribute usage |
655 | =========================== | |
656 | ||
657 | cgroup filesystem supports certain types of extended attributes in its | |
658 | directories and files. The current supported types are: | |
659 | - Trusted (XATTR_TRUSTED) | |
660 | - Security (XATTR_SECURITY) | |
661 | ||
662 | Both require CAP_SYS_ADMIN capability to set. | |
663 | ||
664 | Like in tmpfs, the extended attributes in cgroup filesystem are stored | |
665 | using kernel memory and it's advised to keep the usage at minimum. This | |
666 | is the reason why user defined extended attributes are not supported, since | |
667 | any user can do it and there's no limit in the value size. | |
668 | ||
669 | The current known users for this feature are SELinux to limit cgroup usage | |
670 | in containers and systemd for assorted meta data like main PID in a cgroup | |
671 | (systemd creates a cgroup per service). | |
672 | ||
673 | 5. Questions | |
ddbcc7e8 PM |
674 | ============ |
675 | ||
676 | Q: what's up with this '/bin/echo' ? | |
677 | A: bash's builtin 'echo' command does not check calls to write() against | |
678 | errors. If you use it in the cgroup file system, you won't be | |
679 | able to tell whether a command succeeded or failed. | |
680 | ||
681 | Q: When I attach processes, only the first of the line gets really attached ! | |
682 | A: We can only return one error code per call to write(). So you should also | |
83b061fc | 683 | put only ONE PID. |
ddbcc7e8 | 684 |