blob: 9a5966fb566ccda8349a53b28834ce48f3d7faab [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001#ifndef _LINUX_CGROUP_H
2#define _LINUX_CGROUP_H
3/*
4 * cgroup interface
5 *
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
8 *
9 */
10
11#include <linux/sched.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070012#include <linux/cpumask.h>
13#include <linux/nodemask.h>
Tejun Heoeb6fd502012-11-09 09:12:29 -080014#include <linux/rculist.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070015#include <linux/cgroupstats.h>
Tejun Heo25a7e682013-04-14 20:15:25 -070016#include <linux/fs.h>
Tejun Heo7da11272013-12-05 12:28:04 -050017#include <linux/seq_file.h>
Tejun Heo2bd59d42014-02-11 11:52:49 -050018#include <linux/kernfs.h>
Tejun Heo49d1dc42015-09-18 11:56:28 -040019#include <linux/jump_label.h>
Aditya Kalia79a9082016-01-29 02:54:06 -060020#include <linux/nsproxy.h>
21#include <linux/types.h>
22#include <linux/ns_common.h>
23#include <linux/nsproxy.h>
24#include <linux/user_namespace.h>
Tejun Heob4a04ab2015-05-13 15:38:40 -040025
26#include <linux/cgroup-defs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070027
28#ifdef CONFIG_CGROUPS
29
Tejun Heo6abc8ca2015-08-04 15:20:55 -040030/*
31 * All weight knobs on the default hierarhcy should use the following min,
32 * default and max values. The default value is the logarithmic center of
33 * MIN and MAX and allows 100x to be expressed in both directions.
34 */
35#define CGROUP_WEIGHT_MIN 1
36#define CGROUP_WEIGHT_DFL 100
37#define CGROUP_WEIGHT_MAX 10000
38
Tejun Heoc326aa22015-05-13 16:24:16 -040039/* a css_task_iter should be treated as an opaque object */
40struct css_task_iter {
41 struct cgroup_subsys *ss;
Paul Menageddbcc7e2007-10-18 23:39:30 -070042
Tejun Heoc326aa22015-05-13 16:24:16 -040043 struct list_head *cset_pos;
44 struct list_head *cset_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -070045
Tejun Heoc326aa22015-05-13 16:24:16 -040046 struct list_head *task_pos;
47 struct list_head *tasks_head;
48 struct list_head *mg_tasks_head;
Tejun Heoed27b9f2015-10-15 16:41:52 -040049
50 struct css_set *cur_cset;
51 struct task_struct *cur_task;
52 struct list_head iters_node; /* css_set->task_iters */
Paul Menage817929e2007-10-18 23:39:36 -070053};
Tejun Heoc326aa22015-05-13 16:24:16 -040054
55extern struct cgroup_root cgrp_dfl_root;
56extern struct css_set init_css_set;
57
58#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
59#include <linux/cgroup_subsys.h>
Paul Menage817929e2007-10-18 23:39:36 -070060#undef SUBSYS
61
Tejun Heo49d1dc42015-09-18 11:56:28 -040062#define SUBSYS(_x) \
63 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \
64 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
65#include <linux/cgroup_subsys.h>
66#undef SUBSYS
67
68/**
69 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
70 * @ss: subsystem in question
71 */
72#define cgroup_subsys_enabled(ss) \
73 static_branch_likely(&ss ## _enabled_key)
74
75/**
76 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
77 * @ss: subsystem in question
78 */
79#define cgroup_subsys_on_dfl(ss) \
80 static_branch_likely(&ss ## _on_dfl_key)
81
Tejun Heoc326aa22015-05-13 16:24:16 -040082bool css_has_online_children(struct cgroup_subsys_state *css);
83struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
84struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
85 struct cgroup_subsys *ss);
86struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
87 struct cgroup_subsys *ss);
88
Tejun Heo16af4392015-11-20 15:55:52 -050089struct cgroup *cgroup_get_from_path(const char *path);
Martin KaFai Lau1f3fe7e2016-06-30 10:28:42 -070090struct cgroup *cgroup_get_from_fd(int fd);
Tejun Heo16af4392015-11-20 15:55:52 -050091
Tejun Heoc326aa22015-05-13 16:24:16 -040092int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
93int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
94
95int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
96int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
97int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo34c06252015-11-05 00:12:24 -050098void cgroup_file_notify(struct cgroup_file *cfile);
Tejun Heoc326aa22015-05-13 16:24:16 -040099
Tejun Heo4c737b42016-08-10 11:23:44 -0400100int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
Tejun Heoc326aa22015-05-13 16:24:16 -0400101int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
102int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
103 struct pid *pid, struct task_struct *tsk);
104
105void cgroup_fork(struct task_struct *p);
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500106extern int cgroup_can_fork(struct task_struct *p);
107extern void cgroup_cancel_fork(struct task_struct *p);
108extern void cgroup_post_fork(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400109void cgroup_exit(struct task_struct *p);
Tejun Heo2e91fa72015-10-15 16:41:53 -0400110void cgroup_free(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400111
112int cgroup_init_early(void);
113int cgroup_init(void);
114
Tejun Heo5c9d5352014-05-16 13:22:48 -0400115/*
Tejun Heoc326aa22015-05-13 16:24:16 -0400116 * Iteration helpers and macros.
Tejun Heo5c9d5352014-05-16 13:22:48 -0400117 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700118
Tejun Heoc326aa22015-05-13 16:24:16 -0400119struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
120 struct cgroup_subsys_state *parent);
121struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
122 struct cgroup_subsys_state *css);
123struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
124struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
125 struct cgroup_subsys_state *css);
Tejun Heo72c97e52013-08-08 20:11:22 -0400126
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500127struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
128 struct cgroup_subsys_state **dst_cssp);
129struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
130 struct cgroup_subsys_state **dst_cssp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700131
Tejun Heoc326aa22015-05-13 16:24:16 -0400132void css_task_iter_start(struct cgroup_subsys_state *css,
133 struct css_task_iter *it);
134struct task_struct *css_task_iter_next(struct css_task_iter *it);
135void css_task_iter_end(struct css_task_iter *it);
Tejun Heo0ae78e02013-08-13 11:01:54 -0400136
Tejun Heoc326aa22015-05-13 16:24:16 -0400137/**
138 * css_for_each_child - iterate through children of a css
139 * @pos: the css * to use as the loop cursor
140 * @parent: css whose children to walk
141 *
142 * Walk @parent's children. Must be called under rcu_read_lock().
143 *
144 * If a subsystem synchronizes ->css_online() and the start of iteration, a
145 * css which finished ->css_online() is guaranteed to be visible in the
146 * future iterations and will stay visible until the last reference is put.
147 * A css which hasn't finished ->css_online() or already finished
148 * ->css_offline() may show up during traversal. It's each subsystem's
149 * responsibility to synchronize against on/offlining.
150 *
151 * It is allowed to temporarily drop RCU read lock during iteration. The
152 * caller is responsible for ensuring that @pos remains accessible until
153 * the start of the next iteration by, for example, bumping the css refcnt.
154 */
155#define css_for_each_child(pos, parent) \
156 for ((pos) = css_next_child(NULL, (parent)); (pos); \
157 (pos) = css_next_child((pos), (parent)))
Tejun Heod5c419b2014-05-16 13:22:48 -0400158
Tejun Heoc326aa22015-05-13 16:24:16 -0400159/**
160 * css_for_each_descendant_pre - pre-order walk of a css's descendants
161 * @pos: the css * to use as the loop cursor
162 * @root: css whose descendants to walk
163 *
164 * Walk @root's descendants. @root is included in the iteration and the
165 * first node to be visited. Must be called under rcu_read_lock().
166 *
167 * If a subsystem synchronizes ->css_online() and the start of iteration, a
168 * css which finished ->css_online() is guaranteed to be visible in the
169 * future iterations and will stay visible until the last reference is put.
170 * A css which hasn't finished ->css_online() or already finished
171 * ->css_offline() may show up during traversal. It's each subsystem's
172 * responsibility to synchronize against on/offlining.
173 *
174 * For example, the following guarantees that a descendant can't escape
175 * state updates of its ancestors.
176 *
177 * my_online(@css)
178 * {
179 * Lock @css's parent and @css;
180 * Inherit state from the parent;
181 * Unlock both.
182 * }
183 *
184 * my_update_state(@css)
185 * {
186 * css_for_each_descendant_pre(@pos, @css) {
187 * Lock @pos;
188 * if (@pos == @css)
189 * Update @css's state;
190 * else
191 * Verify @pos is alive and inherit state from its parent;
192 * Unlock @pos;
193 * }
194 * }
195 *
196 * As long as the inheriting step, including checking the parent state, is
197 * enclosed inside @pos locking, double-locking the parent isn't necessary
198 * while inheriting. The state update to the parent is guaranteed to be
199 * visible by walking order and, as long as inheriting operations to the
200 * same @pos are atomic to each other, multiple updates racing each other
201 * still result in the correct state. It's guaranateed that at least one
202 * inheritance happens for any css after the latest update to its parent.
203 *
204 * If checking parent's state requires locking the parent, each inheriting
205 * iteration should lock and unlock both @pos->parent and @pos.
206 *
207 * Alternatively, a subsystem may choose to use a single global lock to
208 * synchronize ->css_online() and ->css_offline() against tree-walking
209 * operations.
210 *
211 * It is allowed to temporarily drop RCU read lock during iteration. The
212 * caller is responsible for ensuring that @pos remains accessible until
213 * the start of the next iteration by, for example, bumping the css refcnt.
214 */
215#define css_for_each_descendant_pre(pos, css) \
216 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
217 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo15a4c832014-05-04 15:09:14 -0400218
Tejun Heoc326aa22015-05-13 16:24:16 -0400219/**
220 * css_for_each_descendant_post - post-order walk of a css's descendants
221 * @pos: the css * to use as the loop cursor
222 * @css: css whose descendants to walk
223 *
224 * Similar to css_for_each_descendant_pre() but performs post-order
225 * traversal instead. @root is included in the iteration and the last
226 * node to be visited.
227 *
228 * If a subsystem synchronizes ->css_online() and the start of iteration, a
229 * css which finished ->css_online() is guaranteed to be visible in the
230 * future iterations and will stay visible until the last reference is put.
231 * A css which hasn't finished ->css_online() or already finished
232 * ->css_offline() may show up during traversal. It's each subsystem's
233 * responsibility to synchronize against on/offlining.
234 *
235 * Note that the walk visibility guarantee example described in pre-order
236 * walk doesn't apply the same to post-order walks.
237 */
238#define css_for_each_descendant_post(pos, css) \
239 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
240 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo48ddbe12012-04-01 12:09:56 -0700241
Tejun Heoc326aa22015-05-13 16:24:16 -0400242/**
243 * cgroup_taskset_for_each - iterate cgroup_taskset
244 * @task: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500245 * @dst_css: the destination css
Tejun Heoc326aa22015-05-13 16:24:16 -0400246 * @tset: taskset to iterate
Tejun Heo4530edd2015-09-11 15:00:19 -0400247 *
248 * @tset may contain multiple tasks and they may belong to multiple
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500249 * processes.
250 *
251 * On the v2 hierarchy, there may be tasks from multiple processes and they
252 * may not share the source or destination csses.
253 *
254 * On traditional hierarchies, when there are multiple tasks in @tset, if a
255 * task of a process is in @tset, all tasks of the process are in @tset.
256 * Also, all are guaranteed to share the same source and destination csses.
Tejun Heo4530edd2015-09-11 15:00:19 -0400257 *
258 * Iteration is not in any specific order.
Tejun Heoc326aa22015-05-13 16:24:16 -0400259 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500260#define cgroup_taskset_for_each(task, dst_css, tset) \
261 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \
262 (task); \
263 (task) = cgroup_taskset_next((tset), &(dst_css)))
Tejun Heo0cb51d72014-05-16 13:22:49 -0400264
Tejun Heo4530edd2015-09-11 15:00:19 -0400265/**
266 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
267 * @leader: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500268 * @dst_css: the destination css
Tejun Heo4530edd2015-09-11 15:00:19 -0400269 * @tset: takset to iterate
270 *
271 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
272 * may not contain any.
273 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500274#define cgroup_taskset_for_each_leader(leader, dst_css, tset) \
275 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \
276 (leader); \
277 (leader) = cgroup_taskset_next((tset), &(dst_css))) \
Tejun Heo4530edd2015-09-11 15:00:19 -0400278 if ((leader) != (leader)->group_leader) \
279 ; \
280 else
281
Tejun Heoc326aa22015-05-13 16:24:16 -0400282/*
283 * Inline functions.
284 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700285
Tejun Heo5de01072013-06-12 21:04:52 -0700286/**
287 * css_get - obtain a reference on the specified css
288 * @css: target css
289 *
290 * The caller must already have a reference.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700291 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700292static inline void css_get(struct cgroup_subsys_state *css)
293{
Tejun Heo3b514d22014-05-16 13:22:47 -0400294 if (!(css->flags & CSS_NO_REF))
295 percpu_ref_get(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700296}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800297
Tejun Heo5de01072013-06-12 21:04:52 -0700298/**
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800299 * css_get_many - obtain references on the specified css
300 * @css: target css
301 * @n: number of references to get
302 *
303 * The caller must already have a reference.
304 */
305static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
306{
307 if (!(css->flags & CSS_NO_REF))
308 percpu_ref_get_many(&css->refcnt, n);
309}
310
311/**
Tejun Heo6f4524d2014-05-16 13:22:52 -0400312 * css_tryget - try to obtain a reference on the specified css
313 * @css: target css
314 *
315 * Obtain a reference on @css unless it already has reached zero and is
316 * being released. This function doesn't care whether @css is on or
317 * offline. The caller naturally needs to ensure that @css is accessible
318 * but doesn't have to be holding a reference on it - IOW, RCU protected
319 * access is good enough for this function. Returns %true if a reference
320 * count was successfully obtained; %false otherwise.
321 */
322static inline bool css_tryget(struct cgroup_subsys_state *css)
323{
324 if (!(css->flags & CSS_NO_REF))
325 return percpu_ref_tryget(&css->refcnt);
326 return true;
327}
328
329/**
Tejun Heoec903c02014-05-13 12:11:01 -0400330 * css_tryget_online - try to obtain a reference on the specified css if online
Tejun Heo5de01072013-06-12 21:04:52 -0700331 * @css: target css
332 *
Tejun Heoec903c02014-05-13 12:11:01 -0400333 * Obtain a reference on @css if it's online. The caller naturally needs
334 * to ensure that @css is accessible but doesn't have to be holding a
Tejun Heo5de01072013-06-12 21:04:52 -0700335 * reference on it - IOW, RCU protected access is good enough for this
336 * function. Returns %true if a reference count was successfully obtained;
337 * %false otherwise.
338 */
Tejun Heoec903c02014-05-13 12:11:01 -0400339static inline bool css_tryget_online(struct cgroup_subsys_state *css)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800340{
Tejun Heo3b514d22014-05-16 13:22:47 -0400341 if (!(css->flags & CSS_NO_REF))
342 return percpu_ref_tryget_live(&css->refcnt);
343 return true;
Paul Menagee7c5ec92009-01-07 18:08:38 -0800344}
345
Tejun Heo5de01072013-06-12 21:04:52 -0700346/**
Tejun Heo829a1ca2017-05-24 12:03:48 -0400347 * css_is_dying - test whether the specified css is dying
348 * @css: target css
349 *
350 * Test whether @css is in the process of offlining or already offline. In
351 * most cases, ->css_online() and ->css_offline() callbacks should be
352 * enough; however, the actual offline operations are RCU delayed and this
353 * test returns %true also when @css is scheduled to be offlined.
354 *
355 * This is useful, for example, when the use case requires synchronous
356 * behavior with respect to cgroup removal. cgroup removal schedules css
357 * offlining but the css can seem alive while the operation is being
358 * delayed. If the delay affects user visible semantics, this test can be
359 * used to resolve the situation.
360 */
361static inline bool css_is_dying(struct cgroup_subsys_state *css)
362{
363 return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt);
364}
365
366/**
Tejun Heo5de01072013-06-12 21:04:52 -0700367 * css_put - put a css reference
368 * @css: target css
369 *
Tejun Heoec903c02014-05-13 12:11:01 -0400370 * Put a reference obtained via css_get() and css_tryget_online().
Tejun Heo5de01072013-06-12 21:04:52 -0700371 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700372static inline void css_put(struct cgroup_subsys_state *css)
373{
Tejun Heo3b514d22014-05-16 13:22:47 -0400374 if (!(css->flags & CSS_NO_REF))
375 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700376}
377
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800378/**
379 * css_put_many - put css references
380 * @css: target css
381 * @n: number of references to put
382 *
383 * Put references obtained via css_get() and css_tryget_online().
384 */
385static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
386{
387 if (!(css->flags & CSS_NO_REF))
388 percpu_ref_put_many(&css->refcnt, n);
389}
390
Tejun Heo86c5c112017-08-11 05:49:01 -0700391static inline void cgroup_get(struct cgroup *cgrp)
392{
393 css_get(&cgrp->self);
394}
395
396static inline bool cgroup_tryget(struct cgroup *cgrp)
397{
398 return css_tryget(&cgrp->self);
399}
400
Tejun Heo16af4392015-11-20 15:55:52 -0500401static inline void cgroup_put(struct cgroup *cgrp)
402{
403 css_put(&cgrp->self);
404}
405
Tejun Heoc326aa22015-05-13 16:24:16 -0400406/**
407 * task_css_set_check - obtain a task's css_set with extra access conditions
408 * @task: the task to obtain css_set for
409 * @__c: extra condition expression to be passed to rcu_dereference_check()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700410 *
Tejun Heoc326aa22015-05-13 16:24:16 -0400411 * A task's css_set is RCU protected, initialized and exited while holding
412 * task_lock(), and can only be modified while holding both cgroup_mutex
413 * and task_lock() while the task is alive. This macro verifies that the
414 * caller is inside proper critical section and returns @task's css_set.
415 *
416 * The caller can also specify additional allowed conditions via @__c, such
417 * as locks used during the cgroup_subsys::attach() methods.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700418 */
Tejun Heoc326aa22015-05-13 16:24:16 -0400419#ifdef CONFIG_PROVE_RCU
420extern struct mutex cgroup_mutex;
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400421extern spinlock_t css_set_lock;
Tejun Heoc326aa22015-05-13 16:24:16 -0400422#define task_css_set_check(task, __c) \
423 rcu_dereference_check((task)->cgroups, \
424 lockdep_is_held(&cgroup_mutex) || \
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400425 lockdep_is_held(&css_set_lock) || \
Tejun Heoc326aa22015-05-13 16:24:16 -0400426 ((task)->flags & PF_EXITING) || (__c))
427#else
428#define task_css_set_check(task, __c) \
429 rcu_dereference((task)->cgroups)
Tejun Heo2bd59d42014-02-11 11:52:49 -0500430#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700431
Tejun Heoc326aa22015-05-13 16:24:16 -0400432/**
433 * task_css_check - obtain css for (task, subsys) w/ extra access conds
434 * @task: the target task
435 * @subsys_id: the target subsystem ID
436 * @__c: extra condition expression to be passed to rcu_dereference_check()
437 *
438 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
439 * synchronization rules are the same as task_css_set_check().
440 */
441#define task_css_check(task, subsys_id, __c) \
442 task_css_set_check((task), (__c))->subsys[(subsys_id)]
443
444/**
445 * task_css_set - obtain a task's css_set
446 * @task: the task to obtain css_set for
447 *
448 * See task_css_set_check().
449 */
450static inline struct css_set *task_css_set(struct task_struct *task)
451{
452 return task_css_set_check(task, false);
453}
454
455/**
456 * task_css - obtain css for (task, subsys)
457 * @task: the target task
458 * @subsys_id: the target subsystem ID
459 *
460 * See task_css_check().
461 */
462static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
463 int subsys_id)
464{
465 return task_css_check(task, subsys_id, false);
466}
467
468/**
Linus Torvaldsbbe179f2015-06-26 19:50:04 -0700469 * task_get_css - find and get the css for (task, subsys)
470 * @task: the target task
471 * @subsys_id: the target subsystem ID
472 *
473 * Find the css for the (@task, @subsys_id) combination, increment a
474 * reference on and return it. This function is guaranteed to return a
475 * valid css.
476 */
477static inline struct cgroup_subsys_state *
478task_get_css(struct task_struct *task, int subsys_id)
479{
480 struct cgroup_subsys_state *css;
481
482 rcu_read_lock();
483 while (true) {
484 css = task_css(task, subsys_id);
485 if (likely(css_tryget_online(css)))
486 break;
487 cpu_relax();
488 }
489 rcu_read_unlock();
490 return css;
491}
492
493/**
Tejun Heoc326aa22015-05-13 16:24:16 -0400494 * task_css_is_root - test whether a task belongs to the root css
495 * @task: the target task
496 * @subsys_id: the target subsystem ID
497 *
498 * Test whether @task belongs to the root css on the specified subsystem.
499 * May be invoked in any context.
500 */
501static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
502{
503 return task_css_check(task, subsys_id, true) ==
504 init_css_set.subsys[subsys_id];
505}
506
507static inline struct cgroup *task_cgroup(struct task_struct *task,
508 int subsys_id)
509{
510 return task_css(task, subsys_id)->cgroup;
511}
Tejun Heoa2dd4242014-03-19 10:23:55 -0400512
Tejun Heo86c5c112017-08-11 05:49:01 -0700513static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
514{
515 return task_css_set(task)->dfl_cgrp;
516}
517
518static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
519{
520 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
521
522 if (parent_css)
523 return container_of(parent_css, struct cgroup, self);
524 return NULL;
525}
526
Tejun Heob11cfb52015-11-20 15:55:52 -0500527/**
528 * cgroup_is_descendant - test ancestry
529 * @cgrp: the cgroup to be tested
530 * @ancestor: possible ancestor of @cgrp
531 *
532 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
533 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
534 * and @ancestor are accessible.
535 */
536static inline bool cgroup_is_descendant(struct cgroup *cgrp,
537 struct cgroup *ancestor)
538{
539 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
540 return false;
541 return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
542}
543
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700544/**
545 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
546 * @task: the task to be tested
547 * @ancestor: possible ancestor of @task's cgroup
548 *
549 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
550 * It follows all the same rules as cgroup_is_descendant, and only applies
551 * to the default hierarchy.
552 */
553static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
554 struct cgroup *ancestor)
555{
556 struct css_set *cset = task_css_set(task);
557
558 return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
559}
560
Tejun Heo07bc3562014-02-13 06:58:39 -0500561/* no synchronization, the result can only be used as a hint */
Tejun Heo27bd4db2015-10-15 16:41:50 -0400562static inline bool cgroup_is_populated(struct cgroup *cgrp)
Tejun Heo07bc3562014-02-13 06:58:39 -0500563{
Tejun Heo27bd4db2015-10-15 16:41:50 -0400564 return cgrp->populated_cnt;
Tejun Heo07bc3562014-02-13 06:58:39 -0500565}
566
Zefan Lif29374b2014-09-19 16:29:31 +0800567/* returns ino associated with a cgroup */
Tejun Heob1664922014-02-11 11:52:49 -0500568static inline ino_t cgroup_ino(struct cgroup *cgrp)
569{
Zefan Lif29374b2014-09-19 16:29:31 +0800570 return cgrp->kn->ino;
Tejun Heob1664922014-02-11 11:52:49 -0500571}
572
Tejun Heob4168642014-05-13 12:16:21 -0400573/* cft/css accessors for cftype->write() operation */
574static inline struct cftype *of_cft(struct kernfs_open_file *of)
Tejun Heo7da11272013-12-05 12:28:04 -0500575{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500576 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500577}
578
Tejun Heob4168642014-05-13 12:16:21 -0400579struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
580
581/* cft/css accessors for cftype->seq_*() operations */
582static inline struct cftype *seq_cft(struct seq_file *seq)
583{
584 return of_cft(seq->private);
585}
586
587static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
588{
589 return of_css(seq->private);
590}
Tejun Heo59f52962014-02-11 11:52:49 -0500591
Tejun Heoe61734c2014-02-12 09:29:50 -0500592/*
593 * Name / path handling functions. All are thin wrappers around the kernfs
594 * counterparts and can be called under any context.
595 */
596
597static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
598{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400599 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500600}
601
Tejun Heo4c737b42016-08-10 11:23:44 -0400602static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
Tejun Heoe61734c2014-02-12 09:29:50 -0500603{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400604 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500605}
606
607static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
608{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400609 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500610}
611
612static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
613{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400614 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500615}
616
Johannes Weinere868a992018-10-26 15:06:31 -0700617static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
618{
619 return &cgrp->psi;
620}
621
Tejun Heof44236a2017-03-16 16:54:24 -0400622static inline void cgroup_init_kthreadd(void)
623{
624 /*
625 * kthreadd is inherited by all kthreads, keep it in the root so
626 * that the new kthreads are guaranteed to stay in the root until
627 * initialization is finished.
628 */
629 current->no_cgroup_migration = 1;
630}
631
632static inline void cgroup_kthread_ready(void)
633{
634 /*
635 * This kthread finished initialization. The creator should have
636 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
637 */
638 current->no_cgroup_migration = 0;
639}
640
Paul Menageddbcc7e2007-10-18 23:39:30 -0700641#else /* !CONFIG_CGROUPS */
642
Tejun Heof3ba5382015-01-06 12:02:46 -0500643struct cgroup_subsys_state;
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700644struct cgroup;
Tejun Heof3ba5382015-01-06 12:02:46 -0500645
Tejun Heoc326aa22015-05-13 16:24:16 -0400646static inline void css_put(struct cgroup_subsys_state *css) {}
647static inline int cgroup_attach_task_all(struct task_struct *from,
648 struct task_struct *t) { return 0; }
649static inline int cgroupstats_build(struct cgroupstats *stats,
650 struct dentry *dentry) { return -EINVAL; }
651
Paul Menageb4f48b62007-10-18 23:39:33 -0700652static inline void cgroup_fork(struct task_struct *p) {}
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500653static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
654static inline void cgroup_cancel_fork(struct task_struct *p) {}
655static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800656static inline void cgroup_exit(struct task_struct *p) {}
Tejun Heo2e91fa72015-10-15 16:41:53 -0400657static inline void cgroup_free(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700658
Tejun Heoc326aa22015-05-13 16:24:16 -0400659static inline int cgroup_init_early(void) { return 0; }
660static inline int cgroup_init(void) { return 0; }
Tejun Heof44236a2017-03-16 16:54:24 -0400661static inline void cgroup_init_kthreadd(void) {}
662static inline void cgroup_kthread_ready(void) {}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200663
Johannes Weinere868a992018-10-26 15:06:31 -0700664static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
665{
666 return NULL;
667}
668
669static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
670{
671 return NULL;
672}
673
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700674static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
675 struct cgroup *ancestor)
676{
677 return true;
678}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700679#endif /* !CONFIG_CGROUPS */
680
Tejun Heobd1060a2015-12-07 17:38:53 -0500681/*
682 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
683 * definition in cgroup-defs.h.
684 */
685#ifdef CONFIG_SOCK_CGROUP_DATA
686
687#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
688extern spinlock_t cgroup_sk_update_lock;
689#endif
690
691void cgroup_sk_alloc_disable(void);
692void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
693void cgroup_sk_free(struct sock_cgroup_data *skcd);
694
695static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
696{
697#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
698 unsigned long v;
699
700 /*
701 * @skcd->val is 64bit but the following is safe on 32bit too as we
702 * just need the lower ulong to be written and read atomically.
703 */
704 v = READ_ONCE(skcd->val);
705
706 if (v & 1)
707 return &cgrp_dfl_root.cgrp;
708
709 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
710#else
711 return (struct cgroup *)(unsigned long)skcd->val;
712#endif
713}
714
715#else /* CONFIG_CGROUP_DATA */
716
717static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
718static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
719
720#endif /* CONFIG_CGROUP_DATA */
721
Aditya Kalia79a9082016-01-29 02:54:06 -0600722struct cgroup_namespace {
723 atomic_t count;
724 struct ns_common ns;
725 struct user_namespace *user_ns;
Eric W. Biedermand08311d2016-08-08 14:25:30 -0500726 struct ucounts *ucounts;
Aditya Kalia79a9082016-01-29 02:54:06 -0600727 struct css_set *root_cset;
728};
729
730extern struct cgroup_namespace init_cgroup_ns;
731
732#ifdef CONFIG_CGROUPS
733
734void free_cgroup_ns(struct cgroup_namespace *ns);
735
736struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
737 struct user_namespace *user_ns,
738 struct cgroup_namespace *old_ns);
739
Tejun Heo4c737b42016-08-10 11:23:44 -0400740int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
741 struct cgroup_namespace *ns);
Aditya Kalia79a9082016-01-29 02:54:06 -0600742
743#else /* !CONFIG_CGROUPS */
744
745static inline void free_cgroup_ns(struct cgroup_namespace *ns) { }
746static inline struct cgroup_namespace *
747copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns,
748 struct cgroup_namespace *old_ns)
749{
750 return old_ns;
751}
752
753#endif /* !CONFIG_CGROUPS */
754
755static inline void get_cgroup_ns(struct cgroup_namespace *ns)
756{
757 if (ns)
758 atomic_inc(&ns->count);
759}
760
761static inline void put_cgroup_ns(struct cgroup_namespace *ns)
762{
763 if (ns && atomic_dec_and_test(&ns->count))
764 free_cgroup_ns(ns);
765}
766
Paul Menageddbcc7e2007-10-18 23:39:30 -0700767#endif /* _LINUX_CGROUP_H */