blob: c0077adeea8334dc136233de439351ca3e742eff [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07002#ifndef _LINUX_CGROUP_H
3#define _LINUX_CGROUP_H
4/*
5 * cgroup interface
6 *
7 * Copyright (C) 2003 BULL SA
8 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
9 *
10 */
11
12#include <linux/sched.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070013#include <linux/cpumask.h>
14#include <linux/nodemask.h>
Tejun Heoeb6fd502012-11-09 09:12:29 -080015#include <linux/rculist.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070016#include <linux/cgroupstats.h>
Tejun Heo25a7e682013-04-14 20:15:25 -070017#include <linux/fs.h>
Tejun Heo7da11272013-12-05 12:28:04 -050018#include <linux/seq_file.h>
Tejun Heo2bd59d42014-02-11 11:52:49 -050019#include <linux/kernfs.h>
Tejun Heo49d1dc42015-09-18 11:56:28 -040020#include <linux/jump_label.h>
Aditya Kalia79a9082016-01-29 02:54:06 -060021#include <linux/types.h>
22#include <linux/ns_common.h>
23#include <linux/nsproxy.h>
24#include <linux/user_namespace.h>
Elena Reshetova387ad962017-02-20 12:19:00 +020025#include <linux/refcount.h>
Tejun Heod2cc5ed2017-09-25 08:12:04 -070026#include <linux/kernel_stat.h>
Tejun Heob4a04ab2015-05-13 15:38:40 -040027
28#include <linux/cgroup-defs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070029
30#ifdef CONFIG_CGROUPS
31
Tejun Heo6abc8ca2015-08-04 15:20:55 -040032/*
33 * All weight knobs on the default hierarhcy should use the following min,
34 * default and max values. The default value is the logarithmic center of
35 * MIN and MAX and allows 100x to be expressed in both directions.
36 */
37#define CGROUP_WEIGHT_MIN 1
38#define CGROUP_WEIGHT_DFL 100
39#define CGROUP_WEIGHT_MAX 10000
40
Tejun Heobc2fb7e2017-05-15 09:34:01 -040041/* walk only threadgroup leaders */
42#define CSS_TASK_ITER_PROCS (1U << 0)
Tejun Heo450ee0c2017-05-15 09:34:03 -040043/* walk all threaded css_sets in the domain */
44#define CSS_TASK_ITER_THREADED (1U << 1)
Tejun Heobc2fb7e2017-05-15 09:34:01 -040045
Tejun Heoc326aa22015-05-13 16:24:16 -040046/* a css_task_iter should be treated as an opaque object */
47struct css_task_iter {
48 struct cgroup_subsys *ss;
Tejun Heobc2fb7e2017-05-15 09:34:01 -040049 unsigned int flags;
Paul Menageddbcc7e2007-10-18 23:39:30 -070050
Tejun Heoc326aa22015-05-13 16:24:16 -040051 struct list_head *cset_pos;
52 struct list_head *cset_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -070053
Tejun Heo450ee0c2017-05-15 09:34:03 -040054 struct list_head *tcset_pos;
55 struct list_head *tcset_head;
56
Tejun Heoc326aa22015-05-13 16:24:16 -040057 struct list_head *task_pos;
58 struct list_head *tasks_head;
59 struct list_head *mg_tasks_head;
Tejun Heoed27b9f2015-10-15 16:41:52 -040060
61 struct css_set *cur_cset;
Tejun Heo450ee0c2017-05-15 09:34:03 -040062 struct css_set *cur_dcset;
Tejun Heoed27b9f2015-10-15 16:41:52 -040063 struct task_struct *cur_task;
64 struct list_head iters_node; /* css_set->task_iters */
Paul Menage817929e2007-10-18 23:39:36 -070065};
Tejun Heoc326aa22015-05-13 16:24:16 -040066
67extern struct cgroup_root cgrp_dfl_root;
68extern struct css_set init_css_set;
69
70#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
71#include <linux/cgroup_subsys.h>
Paul Menage817929e2007-10-18 23:39:36 -070072#undef SUBSYS
73
Tejun Heo49d1dc42015-09-18 11:56:28 -040074#define SUBSYS(_x) \
75 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \
76 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
77#include <linux/cgroup_subsys.h>
78#undef SUBSYS
79
80/**
81 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
82 * @ss: subsystem in question
83 */
84#define cgroup_subsys_enabled(ss) \
85 static_branch_likely(&ss ## _enabled_key)
86
87/**
88 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
89 * @ss: subsystem in question
90 */
91#define cgroup_subsys_on_dfl(ss) \
92 static_branch_likely(&ss ## _on_dfl_key)
93
Tejun Heoc326aa22015-05-13 16:24:16 -040094bool css_has_online_children(struct cgroup_subsys_state *css);
95struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
Dennis Zhoufc5a8282018-12-05 12:10:36 -050096struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup,
97 struct cgroup_subsys *ss);
Tejun Heoc326aa22015-05-13 16:24:16 -040098struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
99 struct cgroup_subsys *ss);
100struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
101 struct cgroup_subsys *ss);
102
Tejun Heo16af4392015-11-20 15:55:52 -0500103struct cgroup *cgroup_get_from_path(const char *path);
Martin KaFai Lau1f3fe7e2016-06-30 10:28:42 -0700104struct cgroup *cgroup_get_from_fd(int fd);
Tejun Heo16af4392015-11-20 15:55:52 -0500105
Tejun Heoc326aa22015-05-13 16:24:16 -0400106int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
107int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
108
109int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
110int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
111int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo34c06252015-11-05 00:12:24 -0500112void cgroup_file_notify(struct cgroup_file *cfile);
Tejun Heoc326aa22015-05-13 16:24:16 -0400113
Tejun Heo4c737b42016-08-10 11:23:44 -0400114int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
Tejun Heoc326aa22015-05-13 16:24:16 -0400115int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
116int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
117 struct pid *pid, struct task_struct *tsk);
118
119void cgroup_fork(struct task_struct *p);
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500120extern int cgroup_can_fork(struct task_struct *p);
121extern void cgroup_cancel_fork(struct task_struct *p);
122extern void cgroup_post_fork(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400123void cgroup_exit(struct task_struct *p);
Oleg Nesterov51bee5a2019-01-28 17:00:13 +0100124void cgroup_release(struct task_struct *p);
Tejun Heo2e91fa72015-10-15 16:41:53 -0400125void cgroup_free(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400126
127int cgroup_init_early(void);
128int cgroup_init(void);
129
Tejun Heo5c9d5352014-05-16 13:22:48 -0400130/*
Tejun Heoc326aa22015-05-13 16:24:16 -0400131 * Iteration helpers and macros.
Tejun Heo5c9d5352014-05-16 13:22:48 -0400132 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700133
Tejun Heoc326aa22015-05-13 16:24:16 -0400134struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
135 struct cgroup_subsys_state *parent);
136struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
137 struct cgroup_subsys_state *css);
138struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
139struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
140 struct cgroup_subsys_state *css);
Tejun Heo72c97e52013-08-08 20:11:22 -0400141
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500142struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
143 struct cgroup_subsys_state **dst_cssp);
144struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
145 struct cgroup_subsys_state **dst_cssp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700146
Tejun Heobc2fb7e2017-05-15 09:34:01 -0400147void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
Tejun Heoc326aa22015-05-13 16:24:16 -0400148 struct css_task_iter *it);
149struct task_struct *css_task_iter_next(struct css_task_iter *it);
150void css_task_iter_end(struct css_task_iter *it);
Tejun Heo0ae78e02013-08-13 11:01:54 -0400151
Tejun Heoc326aa22015-05-13 16:24:16 -0400152/**
153 * css_for_each_child - iterate through children of a css
154 * @pos: the css * to use as the loop cursor
155 * @parent: css whose children to walk
156 *
157 * Walk @parent's children. Must be called under rcu_read_lock().
158 *
159 * If a subsystem synchronizes ->css_online() and the start of iteration, a
160 * css which finished ->css_online() is guaranteed to be visible in the
161 * future iterations and will stay visible until the last reference is put.
162 * A css which hasn't finished ->css_online() or already finished
163 * ->css_offline() may show up during traversal. It's each subsystem's
164 * responsibility to synchronize against on/offlining.
165 *
166 * It is allowed to temporarily drop RCU read lock during iteration. The
167 * caller is responsible for ensuring that @pos remains accessible until
168 * the start of the next iteration by, for example, bumping the css refcnt.
169 */
170#define css_for_each_child(pos, parent) \
171 for ((pos) = css_next_child(NULL, (parent)); (pos); \
172 (pos) = css_next_child((pos), (parent)))
Tejun Heod5c419b2014-05-16 13:22:48 -0400173
Tejun Heoc326aa22015-05-13 16:24:16 -0400174/**
175 * css_for_each_descendant_pre - pre-order walk of a css's descendants
176 * @pos: the css * to use as the loop cursor
177 * @root: css whose descendants to walk
178 *
179 * Walk @root's descendants. @root is included in the iteration and the
180 * first node to be visited. Must be called under rcu_read_lock().
181 *
182 * If a subsystem synchronizes ->css_online() and the start of iteration, a
183 * css which finished ->css_online() is guaranteed to be visible in the
184 * future iterations and will stay visible until the last reference is put.
185 * A css which hasn't finished ->css_online() or already finished
186 * ->css_offline() may show up during traversal. It's each subsystem's
187 * responsibility to synchronize against on/offlining.
188 *
189 * For example, the following guarantees that a descendant can't escape
190 * state updates of its ancestors.
191 *
192 * my_online(@css)
193 * {
194 * Lock @css's parent and @css;
195 * Inherit state from the parent;
196 * Unlock both.
197 * }
198 *
199 * my_update_state(@css)
200 * {
201 * css_for_each_descendant_pre(@pos, @css) {
202 * Lock @pos;
203 * if (@pos == @css)
204 * Update @css's state;
205 * else
206 * Verify @pos is alive and inherit state from its parent;
207 * Unlock @pos;
208 * }
209 * }
210 *
211 * As long as the inheriting step, including checking the parent state, is
212 * enclosed inside @pos locking, double-locking the parent isn't necessary
213 * while inheriting. The state update to the parent is guaranteed to be
214 * visible by walking order and, as long as inheriting operations to the
215 * same @pos are atomic to each other, multiple updates racing each other
216 * still result in the correct state. It's guaranateed that at least one
217 * inheritance happens for any css after the latest update to its parent.
218 *
219 * If checking parent's state requires locking the parent, each inheriting
220 * iteration should lock and unlock both @pos->parent and @pos.
221 *
222 * Alternatively, a subsystem may choose to use a single global lock to
223 * synchronize ->css_online() and ->css_offline() against tree-walking
224 * operations.
225 *
226 * It is allowed to temporarily drop RCU read lock during iteration. The
227 * caller is responsible for ensuring that @pos remains accessible until
228 * the start of the next iteration by, for example, bumping the css refcnt.
229 */
230#define css_for_each_descendant_pre(pos, css) \
231 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
232 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo15a4c832014-05-04 15:09:14 -0400233
Tejun Heoc326aa22015-05-13 16:24:16 -0400234/**
235 * css_for_each_descendant_post - post-order walk of a css's descendants
236 * @pos: the css * to use as the loop cursor
237 * @css: css whose descendants to walk
238 *
239 * Similar to css_for_each_descendant_pre() but performs post-order
240 * traversal instead. @root is included in the iteration and the last
241 * node to be visited.
242 *
243 * If a subsystem synchronizes ->css_online() and the start of iteration, a
244 * css which finished ->css_online() is guaranteed to be visible in the
245 * future iterations and will stay visible until the last reference is put.
246 * A css which hasn't finished ->css_online() or already finished
247 * ->css_offline() may show up during traversal. It's each subsystem's
248 * responsibility to synchronize against on/offlining.
249 *
250 * Note that the walk visibility guarantee example described in pre-order
251 * walk doesn't apply the same to post-order walks.
252 */
253#define css_for_each_descendant_post(pos, css) \
254 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
255 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo48ddbe12012-04-01 12:09:56 -0700256
Tejun Heoc326aa22015-05-13 16:24:16 -0400257/**
258 * cgroup_taskset_for_each - iterate cgroup_taskset
259 * @task: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500260 * @dst_css: the destination css
Tejun Heoc326aa22015-05-13 16:24:16 -0400261 * @tset: taskset to iterate
Tejun Heo4530edd2015-09-11 15:00:19 -0400262 *
263 * @tset may contain multiple tasks and they may belong to multiple
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500264 * processes.
265 *
266 * On the v2 hierarchy, there may be tasks from multiple processes and they
267 * may not share the source or destination csses.
268 *
269 * On traditional hierarchies, when there are multiple tasks in @tset, if a
270 * task of a process is in @tset, all tasks of the process are in @tset.
271 * Also, all are guaranteed to share the same source and destination csses.
Tejun Heo4530edd2015-09-11 15:00:19 -0400272 *
273 * Iteration is not in any specific order.
Tejun Heoc326aa22015-05-13 16:24:16 -0400274 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500275#define cgroup_taskset_for_each(task, dst_css, tset) \
276 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \
277 (task); \
278 (task) = cgroup_taskset_next((tset), &(dst_css)))
Tejun Heo0cb51d72014-05-16 13:22:49 -0400279
Tejun Heo4530edd2015-09-11 15:00:19 -0400280/**
281 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
282 * @leader: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500283 * @dst_css: the destination css
Geliang Tang7b4632f2016-12-24 23:28:35 +0800284 * @tset: taskset to iterate
Tejun Heo4530edd2015-09-11 15:00:19 -0400285 *
286 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
287 * may not contain any.
288 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500289#define cgroup_taskset_for_each_leader(leader, dst_css, tset) \
290 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \
291 (leader); \
292 (leader) = cgroup_taskset_next((tset), &(dst_css))) \
Tejun Heo4530edd2015-09-11 15:00:19 -0400293 if ((leader) != (leader)->group_leader) \
294 ; \
295 else
296
Tejun Heoc326aa22015-05-13 16:24:16 -0400297/*
298 * Inline functions.
299 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700300
Tejun Heo5de01072013-06-12 21:04:52 -0700301/**
302 * css_get - obtain a reference on the specified css
303 * @css: target css
304 *
305 * The caller must already have a reference.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700306 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700307static inline void css_get(struct cgroup_subsys_state *css)
308{
Tejun Heo3b514d22014-05-16 13:22:47 -0400309 if (!(css->flags & CSS_NO_REF))
310 percpu_ref_get(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700311}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800312
Tejun Heo5de01072013-06-12 21:04:52 -0700313/**
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800314 * css_get_many - obtain references on the specified css
315 * @css: target css
316 * @n: number of references to get
317 *
318 * The caller must already have a reference.
319 */
320static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
321{
322 if (!(css->flags & CSS_NO_REF))
323 percpu_ref_get_many(&css->refcnt, n);
324}
325
326/**
Tejun Heo6f4524d2014-05-16 13:22:52 -0400327 * css_tryget - try to obtain a reference on the specified css
328 * @css: target css
329 *
330 * Obtain a reference on @css unless it already has reached zero and is
331 * being released. This function doesn't care whether @css is on or
332 * offline. The caller naturally needs to ensure that @css is accessible
333 * but doesn't have to be holding a reference on it - IOW, RCU protected
334 * access is good enough for this function. Returns %true if a reference
335 * count was successfully obtained; %false otherwise.
336 */
337static inline bool css_tryget(struct cgroup_subsys_state *css)
338{
339 if (!(css->flags & CSS_NO_REF))
340 return percpu_ref_tryget(&css->refcnt);
341 return true;
342}
343
344/**
Tejun Heoec903c02014-05-13 12:11:01 -0400345 * css_tryget_online - try to obtain a reference on the specified css if online
Tejun Heo5de01072013-06-12 21:04:52 -0700346 * @css: target css
347 *
Tejun Heoec903c02014-05-13 12:11:01 -0400348 * Obtain a reference on @css if it's online. The caller naturally needs
349 * to ensure that @css is accessible but doesn't have to be holding a
Tejun Heo5de01072013-06-12 21:04:52 -0700350 * reference on it - IOW, RCU protected access is good enough for this
351 * function. Returns %true if a reference count was successfully obtained;
352 * %false otherwise.
353 */
Tejun Heoec903c02014-05-13 12:11:01 -0400354static inline bool css_tryget_online(struct cgroup_subsys_state *css)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800355{
Tejun Heo3b514d22014-05-16 13:22:47 -0400356 if (!(css->flags & CSS_NO_REF))
357 return percpu_ref_tryget_live(&css->refcnt);
358 return true;
Paul Menagee7c5ec92009-01-07 18:08:38 -0800359}
360
Tejun Heo5de01072013-06-12 21:04:52 -0700361/**
Tejun Heo41c25702017-05-24 12:03:48 -0400362 * css_is_dying - test whether the specified css is dying
363 * @css: target css
364 *
365 * Test whether @css is in the process of offlining or already offline. In
366 * most cases, ->css_online() and ->css_offline() callbacks should be
367 * enough; however, the actual offline operations are RCU delayed and this
368 * test returns %true also when @css is scheduled to be offlined.
369 *
370 * This is useful, for example, when the use case requires synchronous
371 * behavior with respect to cgroup removal. cgroup removal schedules css
372 * offlining but the css can seem alive while the operation is being
373 * delayed. If the delay affects user visible semantics, this test can be
374 * used to resolve the situation.
375 */
376static inline bool css_is_dying(struct cgroup_subsys_state *css)
377{
378 return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt);
379}
380
381/**
Tejun Heo5de01072013-06-12 21:04:52 -0700382 * css_put - put a css reference
383 * @css: target css
384 *
Tejun Heoec903c02014-05-13 12:11:01 -0400385 * Put a reference obtained via css_get() and css_tryget_online().
Tejun Heo5de01072013-06-12 21:04:52 -0700386 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700387static inline void css_put(struct cgroup_subsys_state *css)
388{
Tejun Heo3b514d22014-05-16 13:22:47 -0400389 if (!(css->flags & CSS_NO_REF))
390 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700391}
392
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800393/**
394 * css_put_many - put css references
395 * @css: target css
396 * @n: number of references to put
397 *
398 * Put references obtained via css_get() and css_tryget_online().
399 */
400static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
401{
402 if (!(css->flags & CSS_NO_REF))
403 percpu_ref_put_many(&css->refcnt, n);
404}
405
Tejun Heo3e489302017-08-11 05:49:01 -0700406static inline void cgroup_get(struct cgroup *cgrp)
407{
408 css_get(&cgrp->self);
409}
410
411static inline bool cgroup_tryget(struct cgroup *cgrp)
412{
413 return css_tryget(&cgrp->self);
414}
415
Tejun Heo16af4392015-11-20 15:55:52 -0500416static inline void cgroup_put(struct cgroup *cgrp)
417{
418 css_put(&cgrp->self);
419}
420
Tejun Heoc326aa22015-05-13 16:24:16 -0400421/**
422 * task_css_set_check - obtain a task's css_set with extra access conditions
423 * @task: the task to obtain css_set for
424 * @__c: extra condition expression to be passed to rcu_dereference_check()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700425 *
Tejun Heoc326aa22015-05-13 16:24:16 -0400426 * A task's css_set is RCU protected, initialized and exited while holding
427 * task_lock(), and can only be modified while holding both cgroup_mutex
428 * and task_lock() while the task is alive. This macro verifies that the
429 * caller is inside proper critical section and returns @task's css_set.
430 *
431 * The caller can also specify additional allowed conditions via @__c, such
432 * as locks used during the cgroup_subsys::attach() methods.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700433 */
Tejun Heoc326aa22015-05-13 16:24:16 -0400434#ifdef CONFIG_PROVE_RCU
435extern struct mutex cgroup_mutex;
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400436extern spinlock_t css_set_lock;
Tejun Heoc326aa22015-05-13 16:24:16 -0400437#define task_css_set_check(task, __c) \
438 rcu_dereference_check((task)->cgroups, \
439 lockdep_is_held(&cgroup_mutex) || \
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400440 lockdep_is_held(&css_set_lock) || \
Tejun Heoc326aa22015-05-13 16:24:16 -0400441 ((task)->flags & PF_EXITING) || (__c))
442#else
443#define task_css_set_check(task, __c) \
444 rcu_dereference((task)->cgroups)
Tejun Heo2bd59d42014-02-11 11:52:49 -0500445#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700446
Tejun Heoc326aa22015-05-13 16:24:16 -0400447/**
448 * task_css_check - obtain css for (task, subsys) w/ extra access conds
449 * @task: the target task
450 * @subsys_id: the target subsystem ID
451 * @__c: extra condition expression to be passed to rcu_dereference_check()
452 *
453 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
454 * synchronization rules are the same as task_css_set_check().
455 */
456#define task_css_check(task, subsys_id, __c) \
457 task_css_set_check((task), (__c))->subsys[(subsys_id)]
458
459/**
460 * task_css_set - obtain a task's css_set
461 * @task: the task to obtain css_set for
462 *
463 * See task_css_set_check().
464 */
465static inline struct css_set *task_css_set(struct task_struct *task)
466{
467 return task_css_set_check(task, false);
468}
469
470/**
471 * task_css - obtain css for (task, subsys)
472 * @task: the target task
473 * @subsys_id: the target subsystem ID
474 *
475 * See task_css_check().
476 */
477static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
478 int subsys_id)
479{
480 return task_css_check(task, subsys_id, false);
481}
482
483/**
Linus Torvaldsbbe179f2015-06-26 19:50:04 -0700484 * task_get_css - find and get the css for (task, subsys)
485 * @task: the target task
486 * @subsys_id: the target subsystem ID
487 *
488 * Find the css for the (@task, @subsys_id) combination, increment a
489 * reference on and return it. This function is guaranteed to return a
490 * valid css.
491 */
492static inline struct cgroup_subsys_state *
493task_get_css(struct task_struct *task, int subsys_id)
494{
495 struct cgroup_subsys_state *css;
496
497 rcu_read_lock();
498 while (true) {
499 css = task_css(task, subsys_id);
500 if (likely(css_tryget_online(css)))
501 break;
502 cpu_relax();
503 }
504 rcu_read_unlock();
505 return css;
506}
507
508/**
Tejun Heoc326aa22015-05-13 16:24:16 -0400509 * task_css_is_root - test whether a task belongs to the root css
510 * @task: the target task
511 * @subsys_id: the target subsystem ID
512 *
513 * Test whether @task belongs to the root css on the specified subsystem.
514 * May be invoked in any context.
515 */
516static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
517{
518 return task_css_check(task, subsys_id, true) ==
519 init_css_set.subsys[subsys_id];
520}
521
522static inline struct cgroup *task_cgroup(struct task_struct *task,
523 int subsys_id)
524{
525 return task_css(task, subsys_id)->cgroup;
526}
Tejun Heoa2dd4242014-03-19 10:23:55 -0400527
Tejun Heo3e489302017-08-11 05:49:01 -0700528static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
529{
530 return task_css_set(task)->dfl_cgrp;
531}
532
533static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
534{
535 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
536
537 if (parent_css)
538 return container_of(parent_css, struct cgroup, self);
539 return NULL;
540}
541
Tejun Heob11cfb52015-11-20 15:55:52 -0500542/**
543 * cgroup_is_descendant - test ancestry
544 * @cgrp: the cgroup to be tested
545 * @ancestor: possible ancestor of @cgrp
546 *
547 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
548 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
549 * and @ancestor are accessible.
550 */
551static inline bool cgroup_is_descendant(struct cgroup *cgrp,
552 struct cgroup *ancestor)
553{
554 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
555 return false;
556 return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
557}
558
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700559/**
Andrey Ignatov77236282018-08-12 10:49:27 -0700560 * cgroup_ancestor - find ancestor of cgroup
561 * @cgrp: cgroup to find ancestor of
562 * @ancestor_level: level of ancestor to find starting from root
563 *
564 * Find ancestor of cgroup at specified level starting from root if it exists
565 * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at
566 * @ancestor_level.
567 *
568 * This function is safe to call as long as @cgrp is accessible.
569 */
570static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
571 int ancestor_level)
572{
Andrey Ignatov77236282018-08-12 10:49:27 -0700573 if (cgrp->level < ancestor_level)
574 return NULL;
Andrey Ignatov808c43b2018-09-21 17:03:27 -0700575 while (cgrp && cgrp->level > ancestor_level)
576 cgrp = cgroup_parent(cgrp);
577 return cgrp;
Andrey Ignatov77236282018-08-12 10:49:27 -0700578}
579
580/**
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700581 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
582 * @task: the task to be tested
583 * @ancestor: possible ancestor of @task's cgroup
584 *
585 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
586 * It follows all the same rules as cgroup_is_descendant, and only applies
587 * to the default hierarchy.
588 */
589static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
590 struct cgroup *ancestor)
591{
592 struct css_set *cset = task_css_set(task);
593
594 return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
595}
596
Tejun Heo07bc3562014-02-13 06:58:39 -0500597/* no synchronization, the result can only be used as a hint */
Tejun Heo27bd4db2015-10-15 16:41:50 -0400598static inline bool cgroup_is_populated(struct cgroup *cgrp)
Tejun Heo07bc3562014-02-13 06:58:39 -0500599{
Tejun Heo454000a2017-05-15 09:34:02 -0400600 return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children +
601 cgrp->nr_populated_threaded_children;
Tejun Heo07bc3562014-02-13 06:58:39 -0500602}
603
Zefan Lif29374b2014-09-19 16:29:31 +0800604/* returns ino associated with a cgroup */
Tejun Heob1664922014-02-11 11:52:49 -0500605static inline ino_t cgroup_ino(struct cgroup *cgrp)
606{
Shaohua Lic53cd492017-07-12 11:49:50 -0700607 return cgrp->kn->id.ino;
Tejun Heob1664922014-02-11 11:52:49 -0500608}
609
Tejun Heob4168642014-05-13 12:16:21 -0400610/* cft/css accessors for cftype->write() operation */
611static inline struct cftype *of_cft(struct kernfs_open_file *of)
Tejun Heo7da11272013-12-05 12:28:04 -0500612{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500613 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500614}
615
Tejun Heob4168642014-05-13 12:16:21 -0400616struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
617
618/* cft/css accessors for cftype->seq_*() operations */
619static inline struct cftype *seq_cft(struct seq_file *seq)
620{
621 return of_cft(seq->private);
622}
623
624static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
625{
626 return of_css(seq->private);
627}
Tejun Heo59f52962014-02-11 11:52:49 -0500628
Tejun Heoe61734c2014-02-12 09:29:50 -0500629/*
630 * Name / path handling functions. All are thin wrappers around the kernfs
631 * counterparts and can be called under any context.
632 */
633
634static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
635{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400636 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500637}
638
Tejun Heo4c737b42016-08-10 11:23:44 -0400639static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
Tejun Heoe61734c2014-02-12 09:29:50 -0500640{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400641 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500642}
643
644static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
645{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400646 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500647}
648
649static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
650{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400651 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500652}
653
Johannes Weiner2ce71352018-10-26 15:06:31 -0700654static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
655{
656 return &cgrp->psi;
657}
658
Tejun Heo77f88792017-03-16 16:54:24 -0400659static inline void cgroup_init_kthreadd(void)
660{
661 /*
662 * kthreadd is inherited by all kthreads, keep it in the root so
663 * that the new kthreads are guaranteed to stay in the root until
664 * initialization is finished.
665 */
666 current->no_cgroup_migration = 1;
667}
668
669static inline void cgroup_kthread_ready(void)
670{
671 /*
672 * This kthread finished initialization. The creator should have
673 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
674 */
675 current->no_cgroup_migration = 0;
676}
677
Shaohua Li121508d2017-07-12 11:49:52 -0700678static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
679{
680 return &cgrp->kn->id;
681}
Shaohua Li69fd5c32017-07-12 11:49:55 -0700682
683void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
684 char *buf, size_t buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700685#else /* !CONFIG_CGROUPS */
686
Tejun Heof3ba5382015-01-06 12:02:46 -0500687struct cgroup_subsys_state;
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700688struct cgroup;
Tejun Heof3ba5382015-01-06 12:02:46 -0500689
Tejun Heoc326aa22015-05-13 16:24:16 -0400690static inline void css_put(struct cgroup_subsys_state *css) {}
691static inline int cgroup_attach_task_all(struct task_struct *from,
692 struct task_struct *t) { return 0; }
693static inline int cgroupstats_build(struct cgroupstats *stats,
694 struct dentry *dentry) { return -EINVAL; }
695
Paul Menageb4f48b62007-10-18 23:39:33 -0700696static inline void cgroup_fork(struct task_struct *p) {}
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500697static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
698static inline void cgroup_cancel_fork(struct task_struct *p) {}
699static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800700static inline void cgroup_exit(struct task_struct *p) {}
Oleg Nesterov51bee5a2019-01-28 17:00:13 +0100701static inline void cgroup_release(struct task_struct *p) {}
Tejun Heo2e91fa72015-10-15 16:41:53 -0400702static inline void cgroup_free(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700703
Tejun Heoc326aa22015-05-13 16:24:16 -0400704static inline int cgroup_init_early(void) { return 0; }
705static inline int cgroup_init(void) { return 0; }
Tejun Heo77f88792017-03-16 16:54:24 -0400706static inline void cgroup_init_kthreadd(void) {}
707static inline void cgroup_kthread_ready(void) {}
Shaohua Li121508d2017-07-12 11:49:52 -0700708static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
709{
710 return NULL;
711}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200712
Johannes Weiner2ce71352018-10-26 15:06:31 -0700713static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
714{
715 return NULL;
716}
717
718static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
719{
720 return NULL;
721}
722
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700723static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
724 struct cgroup *ancestor)
725{
726 return true;
727}
Shaohua Li69fd5c32017-07-12 11:49:55 -0700728
729static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
730 char *buf, size_t buflen) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700731#endif /* !CONFIG_CGROUPS */
732
Tejun Heo6162cef2018-04-26 14:29:05 -0700733#ifdef CONFIG_CGROUPS
734/*
735 * cgroup scalable recursive statistics.
736 */
737void cgroup_rstat_updated(struct cgroup *cgrp, int cpu);
738void cgroup_rstat_flush(struct cgroup *cgrp);
Tejun Heo0fa294f2018-04-26 14:29:05 -0700739void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp);
Tejun Heo6162cef2018-04-26 14:29:05 -0700740void cgroup_rstat_flush_hold(struct cgroup *cgrp);
741void cgroup_rstat_flush_release(void);
742
Tejun Heobd1060a2015-12-07 17:38:53 -0500743/*
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700744 * Basic resource stats.
745 */
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700746#ifdef CONFIG_CGROUP_CPUACCT
747void cpuacct_charge(struct task_struct *tsk, u64 cputime);
748void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);
749#else
750static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
751static inline void cpuacct_account_field(struct task_struct *tsk, int index,
752 u64 val) {}
753#endif
754
Tejun Heo041cd642017-09-25 08:12:05 -0700755void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec);
756void __cgroup_account_cputime_field(struct cgroup *cgrp,
757 enum cpu_usage_stat index, u64 delta_exec);
758
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700759static inline void cgroup_account_cputime(struct task_struct *task,
760 u64 delta_exec)
761{
Tejun Heo041cd642017-09-25 08:12:05 -0700762 struct cgroup *cgrp;
763
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700764 cpuacct_charge(task, delta_exec);
Tejun Heo041cd642017-09-25 08:12:05 -0700765
766 rcu_read_lock();
767 cgrp = task_dfl_cgroup(task);
768 if (cgroup_parent(cgrp))
769 __cgroup_account_cputime(cgrp, delta_exec);
770 rcu_read_unlock();
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700771}
772
773static inline void cgroup_account_cputime_field(struct task_struct *task,
774 enum cpu_usage_stat index,
775 u64 delta_exec)
776{
Tejun Heo041cd642017-09-25 08:12:05 -0700777 struct cgroup *cgrp;
778
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700779 cpuacct_account_field(task, index, delta_exec);
Tejun Heo041cd642017-09-25 08:12:05 -0700780
781 rcu_read_lock();
782 cgrp = task_dfl_cgroup(task);
783 if (cgroup_parent(cgrp))
784 __cgroup_account_cputime_field(cgrp, index, delta_exec);
785 rcu_read_unlock();
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700786}
787
788#else /* CONFIG_CGROUPS */
789
790static inline void cgroup_account_cputime(struct task_struct *task,
791 u64 delta_exec) {}
792static inline void cgroup_account_cputime_field(struct task_struct *task,
793 enum cpu_usage_stat index,
794 u64 delta_exec) {}
795
796#endif /* CONFIG_CGROUPS */
797
798/*
Tejun Heobd1060a2015-12-07 17:38:53 -0500799 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
800 * definition in cgroup-defs.h.
801 */
802#ifdef CONFIG_SOCK_CGROUP_DATA
803
804#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
805extern spinlock_t cgroup_sk_update_lock;
806#endif
807
808void cgroup_sk_alloc_disable(void);
809void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
810void cgroup_sk_free(struct sock_cgroup_data *skcd);
811
812static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
813{
814#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
815 unsigned long v;
816
817 /*
818 * @skcd->val is 64bit but the following is safe on 32bit too as we
819 * just need the lower ulong to be written and read atomically.
820 */
821 v = READ_ONCE(skcd->val);
822
823 if (v & 1)
824 return &cgrp_dfl_root.cgrp;
825
826 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
827#else
828 return (struct cgroup *)(unsigned long)skcd->val;
829#endif
830}
831
832#else /* CONFIG_CGROUP_DATA */
833
834static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
835static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
836
837#endif /* CONFIG_CGROUP_DATA */
838
Aditya Kalia79a9082016-01-29 02:54:06 -0600839struct cgroup_namespace {
Elena Reshetova387ad962017-02-20 12:19:00 +0200840 refcount_t count;
Aditya Kalia79a9082016-01-29 02:54:06 -0600841 struct ns_common ns;
842 struct user_namespace *user_ns;
Eric W. Biedermand08311d2016-08-08 14:25:30 -0500843 struct ucounts *ucounts;
Aditya Kalia79a9082016-01-29 02:54:06 -0600844 struct css_set *root_cset;
845};
846
847extern struct cgroup_namespace init_cgroup_ns;
848
849#ifdef CONFIG_CGROUPS
850
851void free_cgroup_ns(struct cgroup_namespace *ns);
852
853struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
854 struct user_namespace *user_ns,
855 struct cgroup_namespace *old_ns);
856
Tejun Heo4c737b42016-08-10 11:23:44 -0400857int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
858 struct cgroup_namespace *ns);
Aditya Kalia79a9082016-01-29 02:54:06 -0600859
860#else /* !CONFIG_CGROUPS */
861
862static inline void free_cgroup_ns(struct cgroup_namespace *ns) { }
863static inline struct cgroup_namespace *
864copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns,
865 struct cgroup_namespace *old_ns)
866{
867 return old_ns;
868}
869
870#endif /* !CONFIG_CGROUPS */
871
872static inline void get_cgroup_ns(struct cgroup_namespace *ns)
873{
874 if (ns)
Elena Reshetova387ad962017-02-20 12:19:00 +0200875 refcount_inc(&ns->count);
Aditya Kalia79a9082016-01-29 02:54:06 -0600876}
877
878static inline void put_cgroup_ns(struct cgroup_namespace *ns)
879{
Elena Reshetova387ad962017-02-20 12:19:00 +0200880 if (ns && refcount_dec_and_test(&ns->count))
Aditya Kalia79a9082016-01-29 02:54:06 -0600881 free_cgroup_ns(ns);
882}
883
Roman Gushchin76f969e2019-04-19 10:03:04 -0700884#ifdef CONFIG_CGROUPS
885
886void cgroup_enter_frozen(void);
887void cgroup_leave_frozen(bool always_leave);
888void cgroup_update_frozen(struct cgroup *cgrp);
889void cgroup_freeze(struct cgroup *cgrp, bool freeze);
890void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src,
891 struct cgroup *dst);
Roman Gushchin96b9c592019-04-26 10:59:45 -0700892
Roman Gushchin76f969e2019-04-19 10:03:04 -0700893static inline bool cgroup_task_freeze(struct task_struct *task)
894{
895 bool ret;
896
897 if (task->flags & PF_KTHREAD)
898 return false;
899
900 rcu_read_lock();
901 ret = test_bit(CGRP_FREEZE, &task_dfl_cgroup(task)->flags);
902 rcu_read_unlock();
903
904 return ret;
905}
906
907static inline bool cgroup_task_frozen(struct task_struct *task)
908{
909 return task->frozen;
910}
911
912#else /* !CONFIG_CGROUPS */
913
914static inline void cgroup_enter_frozen(void) { }
915static inline void cgroup_leave_frozen(bool always_leave) { }
916static inline bool cgroup_task_freeze(struct task_struct *task)
917{
918 return false;
919}
920static inline bool cgroup_task_frozen(struct task_struct *task)
921{
922 return false;
923}
924
925#endif /* !CONFIG_CGROUPS */
926
Paul Menageddbcc7e2007-10-18 23:39:30 -0700927#endif /* _LINUX_CGROUP_H */