blob: 0297f930a56e1b08a46141a1d535b7d214183aca [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 Heob636fd32019-05-31 10:38:58 -070046/* internal flags */
47#define CSS_TASK_ITER_SKIPPED (1U << 16)
48
Tejun Heoc326aa22015-05-13 16:24:16 -040049/* a css_task_iter should be treated as an opaque object */
50struct css_task_iter {
51 struct cgroup_subsys *ss;
Tejun Heobc2fb7e2017-05-15 09:34:01 -040052 unsigned int flags;
Paul Menageddbcc7e2007-10-18 23:39:30 -070053
Tejun Heoc326aa22015-05-13 16:24:16 -040054 struct list_head *cset_pos;
55 struct list_head *cset_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -070056
Tejun Heo450ee0c2017-05-15 09:34:03 -040057 struct list_head *tcset_pos;
58 struct list_head *tcset_head;
59
Tejun Heoc326aa22015-05-13 16:24:16 -040060 struct list_head *task_pos;
61 struct list_head *tasks_head;
62 struct list_head *mg_tasks_head;
Tejun Heoc03cd772019-05-31 10:38:58 -070063 struct list_head *dying_tasks_head;
Tejun Heoed27b9f2015-10-15 16:41:52 -040064
65 struct css_set *cur_cset;
Tejun Heo450ee0c2017-05-15 09:34:03 -040066 struct css_set *cur_dcset;
Tejun Heoed27b9f2015-10-15 16:41:52 -040067 struct task_struct *cur_task;
68 struct list_head iters_node; /* css_set->task_iters */
Paul Menage817929e2007-10-18 23:39:36 -070069};
Tejun Heoc326aa22015-05-13 16:24:16 -040070
71extern struct cgroup_root cgrp_dfl_root;
72extern struct css_set init_css_set;
73
74#define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
75#include <linux/cgroup_subsys.h>
Paul Menage817929e2007-10-18 23:39:36 -070076#undef SUBSYS
77
Tejun Heo49d1dc42015-09-18 11:56:28 -040078#define SUBSYS(_x) \
79 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \
80 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
81#include <linux/cgroup_subsys.h>
82#undef SUBSYS
83
84/**
85 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
86 * @ss: subsystem in question
87 */
88#define cgroup_subsys_enabled(ss) \
89 static_branch_likely(&ss ## _enabled_key)
90
91/**
92 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
93 * @ss: subsystem in question
94 */
95#define cgroup_subsys_on_dfl(ss) \
96 static_branch_likely(&ss ## _on_dfl_key)
97
Tejun Heoc326aa22015-05-13 16:24:16 -040098bool css_has_online_children(struct cgroup_subsys_state *css);
99struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
Dennis Zhoufc5a8282018-12-05 12:10:36 -0500100struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup,
101 struct cgroup_subsys *ss);
Tejun Heoc326aa22015-05-13 16:24:16 -0400102struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
103 struct cgroup_subsys *ss);
104struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
105 struct cgroup_subsys *ss);
106
Tejun Heo16af4392015-11-20 15:55:52 -0500107struct cgroup *cgroup_get_from_path(const char *path);
Martin KaFai Lau1f3fe7e2016-06-30 10:28:42 -0700108struct cgroup *cgroup_get_from_fd(int fd);
Tejun Heo16af4392015-11-20 15:55:52 -0500109
Tejun Heoc326aa22015-05-13 16:24:16 -0400110int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
111int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
112
113int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
114int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
115int cgroup_rm_cftypes(struct cftype *cfts);
Tejun Heo34c06252015-11-05 00:12:24 -0500116void cgroup_file_notify(struct cgroup_file *cfile);
Tejun Heoc326aa22015-05-13 16:24:16 -0400117
Tejun Heo4c737b42016-08-10 11:23:44 -0400118int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
Tejun Heoc326aa22015-05-13 16:24:16 -0400119int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
120int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
121 struct pid *pid, struct task_struct *tsk);
122
123void cgroup_fork(struct task_struct *p);
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500124extern int cgroup_can_fork(struct task_struct *p);
125extern void cgroup_cancel_fork(struct task_struct *p);
126extern void cgroup_post_fork(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400127void cgroup_exit(struct task_struct *p);
Oleg Nesterov51bee5a2019-01-28 17:00:13 +0100128void cgroup_release(struct task_struct *p);
Tejun Heo2e91fa72015-10-15 16:41:53 -0400129void cgroup_free(struct task_struct *p);
Tejun Heoc326aa22015-05-13 16:24:16 -0400130
131int cgroup_init_early(void);
132int cgroup_init(void);
133
Tejun Heo5c9d5352014-05-16 13:22:48 -0400134/*
Tejun Heoc326aa22015-05-13 16:24:16 -0400135 * Iteration helpers and macros.
Tejun Heo5c9d5352014-05-16 13:22:48 -0400136 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700137
Tejun Heoc326aa22015-05-13 16:24:16 -0400138struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
139 struct cgroup_subsys_state *parent);
140struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
141 struct cgroup_subsys_state *css);
142struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
143struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
144 struct cgroup_subsys_state *css);
Tejun Heo72c97e52013-08-08 20:11:22 -0400145
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500146struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
147 struct cgroup_subsys_state **dst_cssp);
148struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
149 struct cgroup_subsys_state **dst_cssp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700150
Tejun Heobc2fb7e2017-05-15 09:34:01 -0400151void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
Tejun Heoc326aa22015-05-13 16:24:16 -0400152 struct css_task_iter *it);
153struct task_struct *css_task_iter_next(struct css_task_iter *it);
154void css_task_iter_end(struct css_task_iter *it);
Tejun Heo0ae78e02013-08-13 11:01:54 -0400155
Tejun Heoc326aa22015-05-13 16:24:16 -0400156/**
157 * css_for_each_child - iterate through children of a css
158 * @pos: the css * to use as the loop cursor
159 * @parent: css whose children to walk
160 *
161 * Walk @parent's children. Must be called under rcu_read_lock().
162 *
163 * If a subsystem synchronizes ->css_online() and the start of iteration, a
164 * css which finished ->css_online() is guaranteed to be visible in the
165 * future iterations and will stay visible until the last reference is put.
166 * A css which hasn't finished ->css_online() or already finished
167 * ->css_offline() may show up during traversal. It's each subsystem's
168 * responsibility to synchronize against on/offlining.
169 *
170 * It is allowed to temporarily drop RCU read lock during iteration. The
171 * caller is responsible for ensuring that @pos remains accessible until
172 * the start of the next iteration by, for example, bumping the css refcnt.
173 */
174#define css_for_each_child(pos, parent) \
175 for ((pos) = css_next_child(NULL, (parent)); (pos); \
176 (pos) = css_next_child((pos), (parent)))
Tejun Heod5c419b2014-05-16 13:22:48 -0400177
Tejun Heoc326aa22015-05-13 16:24:16 -0400178/**
179 * css_for_each_descendant_pre - pre-order walk of a css's descendants
180 * @pos: the css * to use as the loop cursor
181 * @root: css whose descendants to walk
182 *
183 * Walk @root's descendants. @root is included in the iteration and the
184 * first node to be visited. Must be called under rcu_read_lock().
185 *
186 * If a subsystem synchronizes ->css_online() and the start of iteration, a
187 * css which finished ->css_online() is guaranteed to be visible in the
188 * future iterations and will stay visible until the last reference is put.
189 * A css which hasn't finished ->css_online() or already finished
190 * ->css_offline() may show up during traversal. It's each subsystem's
191 * responsibility to synchronize against on/offlining.
192 *
193 * For example, the following guarantees that a descendant can't escape
194 * state updates of its ancestors.
195 *
196 * my_online(@css)
197 * {
198 * Lock @css's parent and @css;
199 * Inherit state from the parent;
200 * Unlock both.
201 * }
202 *
203 * my_update_state(@css)
204 * {
205 * css_for_each_descendant_pre(@pos, @css) {
206 * Lock @pos;
207 * if (@pos == @css)
208 * Update @css's state;
209 * else
210 * Verify @pos is alive and inherit state from its parent;
211 * Unlock @pos;
212 * }
213 * }
214 *
215 * As long as the inheriting step, including checking the parent state, is
216 * enclosed inside @pos locking, double-locking the parent isn't necessary
217 * while inheriting. The state update to the parent is guaranteed to be
218 * visible by walking order and, as long as inheriting operations to the
219 * same @pos are atomic to each other, multiple updates racing each other
220 * still result in the correct state. It's guaranateed that at least one
221 * inheritance happens for any css after the latest update to its parent.
222 *
223 * If checking parent's state requires locking the parent, each inheriting
224 * iteration should lock and unlock both @pos->parent and @pos.
225 *
226 * Alternatively, a subsystem may choose to use a single global lock to
227 * synchronize ->css_online() and ->css_offline() against tree-walking
228 * operations.
229 *
230 * It is allowed to temporarily drop RCU read lock during iteration. The
231 * caller is responsible for ensuring that @pos remains accessible until
232 * the start of the next iteration by, for example, bumping the css refcnt.
233 */
234#define css_for_each_descendant_pre(pos, css) \
235 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
236 (pos) = css_next_descendant_pre((pos), (css)))
Tejun Heo15a4c832014-05-04 15:09:14 -0400237
Tejun Heoc326aa22015-05-13 16:24:16 -0400238/**
239 * css_for_each_descendant_post - post-order walk of a css's descendants
240 * @pos: the css * to use as the loop cursor
241 * @css: css whose descendants to walk
242 *
243 * Similar to css_for_each_descendant_pre() but performs post-order
244 * traversal instead. @root is included in the iteration and the last
245 * node to be visited.
246 *
247 * If a subsystem synchronizes ->css_online() and the start of iteration, a
248 * css which finished ->css_online() is guaranteed to be visible in the
249 * future iterations and will stay visible until the last reference is put.
250 * A css which hasn't finished ->css_online() or already finished
251 * ->css_offline() may show up during traversal. It's each subsystem's
252 * responsibility to synchronize against on/offlining.
253 *
254 * Note that the walk visibility guarantee example described in pre-order
255 * walk doesn't apply the same to post-order walks.
256 */
257#define css_for_each_descendant_post(pos, css) \
258 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
259 (pos) = css_next_descendant_post((pos), (css)))
Tejun Heo48ddbe12012-04-01 12:09:56 -0700260
Tejun Heoc326aa22015-05-13 16:24:16 -0400261/**
262 * cgroup_taskset_for_each - iterate cgroup_taskset
263 * @task: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500264 * @dst_css: the destination css
Tejun Heoc326aa22015-05-13 16:24:16 -0400265 * @tset: taskset to iterate
Tejun Heo4530edd2015-09-11 15:00:19 -0400266 *
267 * @tset may contain multiple tasks and they may belong to multiple
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500268 * processes.
269 *
270 * On the v2 hierarchy, there may be tasks from multiple processes and they
271 * may not share the source or destination csses.
272 *
273 * On traditional hierarchies, when there are multiple tasks in @tset, if a
274 * task of a process is in @tset, all tasks of the process are in @tset.
275 * Also, all are guaranteed to share the same source and destination csses.
Tejun Heo4530edd2015-09-11 15:00:19 -0400276 *
277 * Iteration is not in any specific order.
Tejun Heoc326aa22015-05-13 16:24:16 -0400278 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500279#define cgroup_taskset_for_each(task, dst_css, tset) \
280 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \
281 (task); \
282 (task) = cgroup_taskset_next((tset), &(dst_css)))
Tejun Heo0cb51d72014-05-16 13:22:49 -0400283
Tejun Heo4530edd2015-09-11 15:00:19 -0400284/**
285 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
286 * @leader: the loop cursor
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500287 * @dst_css: the destination css
Geliang Tang7b4632f2016-12-24 23:28:35 +0800288 * @tset: taskset to iterate
Tejun Heo4530edd2015-09-11 15:00:19 -0400289 *
290 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
291 * may not contain any.
292 */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -0500293#define cgroup_taskset_for_each_leader(leader, dst_css, tset) \
294 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \
295 (leader); \
296 (leader) = cgroup_taskset_next((tset), &(dst_css))) \
Tejun Heo4530edd2015-09-11 15:00:19 -0400297 if ((leader) != (leader)->group_leader) \
298 ; \
299 else
300
Tejun Heoc326aa22015-05-13 16:24:16 -0400301/*
302 * Inline functions.
303 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700304
Tejun Heo5de01072013-06-12 21:04:52 -0700305/**
306 * css_get - obtain a reference on the specified css
307 * @css: target css
308 *
309 * The caller must already have a reference.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700310 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700311static inline void css_get(struct cgroup_subsys_state *css)
312{
Tejun Heo3b514d22014-05-16 13:22:47 -0400313 if (!(css->flags & CSS_NO_REF))
314 percpu_ref_get(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700315}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800316
Tejun Heo5de01072013-06-12 21:04:52 -0700317/**
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800318 * css_get_many - obtain references on the specified css
319 * @css: target css
320 * @n: number of references to get
321 *
322 * The caller must already have a reference.
323 */
324static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
325{
326 if (!(css->flags & CSS_NO_REF))
327 percpu_ref_get_many(&css->refcnt, n);
328}
329
330/**
Tejun Heo6f4524d2014-05-16 13:22:52 -0400331 * css_tryget - try to obtain a reference on the specified css
332 * @css: target css
333 *
334 * Obtain a reference on @css unless it already has reached zero and is
335 * being released. This function doesn't care whether @css is on or
336 * offline. The caller naturally needs to ensure that @css is accessible
337 * but doesn't have to be holding a reference on it - IOW, RCU protected
338 * access is good enough for this function. Returns %true if a reference
339 * count was successfully obtained; %false otherwise.
340 */
341static inline bool css_tryget(struct cgroup_subsys_state *css)
342{
343 if (!(css->flags & CSS_NO_REF))
344 return percpu_ref_tryget(&css->refcnt);
345 return true;
346}
347
348/**
Tejun Heoec903c02014-05-13 12:11:01 -0400349 * css_tryget_online - try to obtain a reference on the specified css if online
Tejun Heo5de01072013-06-12 21:04:52 -0700350 * @css: target css
351 *
Tejun Heoec903c02014-05-13 12:11:01 -0400352 * Obtain a reference on @css if it's online. The caller naturally needs
353 * to ensure that @css is accessible but doesn't have to be holding a
Tejun Heo5de01072013-06-12 21:04:52 -0700354 * reference on it - IOW, RCU protected access is good enough for this
355 * function. Returns %true if a reference count was successfully obtained;
356 * %false otherwise.
357 */
Tejun Heoec903c02014-05-13 12:11:01 -0400358static inline bool css_tryget_online(struct cgroup_subsys_state *css)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800359{
Tejun Heo3b514d22014-05-16 13:22:47 -0400360 if (!(css->flags & CSS_NO_REF))
361 return percpu_ref_tryget_live(&css->refcnt);
362 return true;
Paul Menagee7c5ec92009-01-07 18:08:38 -0800363}
364
Tejun Heo5de01072013-06-12 21:04:52 -0700365/**
Tejun Heo41c25702017-05-24 12:03:48 -0400366 * css_is_dying - test whether the specified css is dying
367 * @css: target css
368 *
369 * Test whether @css is in the process of offlining or already offline. In
370 * most cases, ->css_online() and ->css_offline() callbacks should be
371 * enough; however, the actual offline operations are RCU delayed and this
372 * test returns %true also when @css is scheduled to be offlined.
373 *
374 * This is useful, for example, when the use case requires synchronous
375 * behavior with respect to cgroup removal. cgroup removal schedules css
376 * offlining but the css can seem alive while the operation is being
377 * delayed. If the delay affects user visible semantics, this test can be
378 * used to resolve the situation.
379 */
380static inline bool css_is_dying(struct cgroup_subsys_state *css)
381{
382 return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt);
383}
384
385/**
Tejun Heo5de01072013-06-12 21:04:52 -0700386 * css_put - put a css reference
387 * @css: target css
388 *
Tejun Heoec903c02014-05-13 12:11:01 -0400389 * Put a reference obtained via css_get() and css_tryget_online().
Tejun Heo5de01072013-06-12 21:04:52 -0700390 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700391static inline void css_put(struct cgroup_subsys_state *css)
392{
Tejun Heo3b514d22014-05-16 13:22:47 -0400393 if (!(css->flags & CSS_NO_REF))
394 percpu_ref_put(&css->refcnt);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700395}
396
Johannes Weinere8ea14c2014-12-10 15:42:42 -0800397/**
398 * css_put_many - put css references
399 * @css: target css
400 * @n: number of references to put
401 *
402 * Put references obtained via css_get() and css_tryget_online().
403 */
404static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
405{
406 if (!(css->flags & CSS_NO_REF))
407 percpu_ref_put_many(&css->refcnt, n);
408}
409
Tejun Heo3e489302017-08-11 05:49:01 -0700410static inline void cgroup_get(struct cgroup *cgrp)
411{
412 css_get(&cgrp->self);
413}
414
415static inline bool cgroup_tryget(struct cgroup *cgrp)
416{
417 return css_tryget(&cgrp->self);
418}
419
Tejun Heo16af4392015-11-20 15:55:52 -0500420static inline void cgroup_put(struct cgroup *cgrp)
421{
422 css_put(&cgrp->self);
423}
424
Tejun Heoc326aa22015-05-13 16:24:16 -0400425/**
426 * task_css_set_check - obtain a task's css_set with extra access conditions
427 * @task: the task to obtain css_set for
428 * @__c: extra condition expression to be passed to rcu_dereference_check()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700429 *
Tejun Heoc326aa22015-05-13 16:24:16 -0400430 * A task's css_set is RCU protected, initialized and exited while holding
431 * task_lock(), and can only be modified while holding both cgroup_mutex
432 * and task_lock() while the task is alive. This macro verifies that the
433 * caller is inside proper critical section and returns @task's css_set.
434 *
435 * The caller can also specify additional allowed conditions via @__c, such
436 * as locks used during the cgroup_subsys::attach() methods.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700437 */
Tejun Heoc326aa22015-05-13 16:24:16 -0400438#ifdef CONFIG_PROVE_RCU
439extern struct mutex cgroup_mutex;
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400440extern spinlock_t css_set_lock;
Tejun Heoc326aa22015-05-13 16:24:16 -0400441#define task_css_set_check(task, __c) \
442 rcu_dereference_check((task)->cgroups, \
443 lockdep_is_held(&cgroup_mutex) || \
Tejun Heof0d9a5f2015-10-15 16:41:53 -0400444 lockdep_is_held(&css_set_lock) || \
Tejun Heoc326aa22015-05-13 16:24:16 -0400445 ((task)->flags & PF_EXITING) || (__c))
446#else
447#define task_css_set_check(task, __c) \
448 rcu_dereference((task)->cgroups)
Tejun Heo2bd59d42014-02-11 11:52:49 -0500449#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700450
Tejun Heoc326aa22015-05-13 16:24:16 -0400451/**
452 * task_css_check - obtain css for (task, subsys) w/ extra access conds
453 * @task: the target task
454 * @subsys_id: the target subsystem ID
455 * @__c: extra condition expression to be passed to rcu_dereference_check()
456 *
457 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
458 * synchronization rules are the same as task_css_set_check().
459 */
460#define task_css_check(task, subsys_id, __c) \
461 task_css_set_check((task), (__c))->subsys[(subsys_id)]
462
463/**
464 * task_css_set - obtain a task's css_set
465 * @task: the task to obtain css_set for
466 *
467 * See task_css_set_check().
468 */
469static inline struct css_set *task_css_set(struct task_struct *task)
470{
471 return task_css_set_check(task, false);
472}
473
474/**
475 * task_css - obtain css for (task, subsys)
476 * @task: the target task
477 * @subsys_id: the target subsystem ID
478 *
479 * See task_css_check().
480 */
481static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
482 int subsys_id)
483{
484 return task_css_check(task, subsys_id, false);
485}
486
487/**
Linus Torvaldsbbe179f2015-06-26 19:50:04 -0700488 * task_get_css - find and get the css for (task, subsys)
489 * @task: the target task
490 * @subsys_id: the target subsystem ID
491 *
492 * Find the css for the (@task, @subsys_id) combination, increment a
493 * reference on and return it. This function is guaranteed to return a
Tejun Heo18fa84a2019-05-29 13:46:25 -0700494 * valid css. The returned css may already have been offlined.
Linus Torvaldsbbe179f2015-06-26 19:50:04 -0700495 */
496static inline struct cgroup_subsys_state *
497task_get_css(struct task_struct *task, int subsys_id)
498{
499 struct cgroup_subsys_state *css;
500
501 rcu_read_lock();
502 while (true) {
503 css = task_css(task, subsys_id);
Tejun Heo18fa84a2019-05-29 13:46:25 -0700504 /*
505 * Can't use css_tryget_online() here. A task which has
506 * PF_EXITING set may stay associated with an offline css.
507 * If such task calls this function, css_tryget_online()
508 * will keep failing.
509 */
510 if (likely(css_tryget(css)))
Linus Torvaldsbbe179f2015-06-26 19:50:04 -0700511 break;
512 cpu_relax();
513 }
514 rcu_read_unlock();
515 return css;
516}
517
518/**
Tejun Heoc326aa22015-05-13 16:24:16 -0400519 * task_css_is_root - test whether a task belongs to the root css
520 * @task: the target task
521 * @subsys_id: the target subsystem ID
522 *
523 * Test whether @task belongs to the root css on the specified subsystem.
524 * May be invoked in any context.
525 */
526static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
527{
528 return task_css_check(task, subsys_id, true) ==
529 init_css_set.subsys[subsys_id];
530}
531
532static inline struct cgroup *task_cgroup(struct task_struct *task,
533 int subsys_id)
534{
535 return task_css(task, subsys_id)->cgroup;
536}
Tejun Heoa2dd4242014-03-19 10:23:55 -0400537
Tejun Heo3e489302017-08-11 05:49:01 -0700538static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
539{
540 return task_css_set(task)->dfl_cgrp;
541}
542
543static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
544{
545 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
546
547 if (parent_css)
548 return container_of(parent_css, struct cgroup, self);
549 return NULL;
550}
551
Tejun Heob11cfb52015-11-20 15:55:52 -0500552/**
553 * cgroup_is_descendant - test ancestry
554 * @cgrp: the cgroup to be tested
555 * @ancestor: possible ancestor of @cgrp
556 *
557 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
558 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
559 * and @ancestor are accessible.
560 */
561static inline bool cgroup_is_descendant(struct cgroup *cgrp,
562 struct cgroup *ancestor)
563{
564 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
565 return false;
566 return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
567}
568
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700569/**
Andrey Ignatov77236282018-08-12 10:49:27 -0700570 * cgroup_ancestor - find ancestor of cgroup
571 * @cgrp: cgroup to find ancestor of
572 * @ancestor_level: level of ancestor to find starting from root
573 *
574 * Find ancestor of cgroup at specified level starting from root if it exists
575 * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at
576 * @ancestor_level.
577 *
578 * This function is safe to call as long as @cgrp is accessible.
579 */
580static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
581 int ancestor_level)
582{
Andrey Ignatov77236282018-08-12 10:49:27 -0700583 if (cgrp->level < ancestor_level)
584 return NULL;
Andrey Ignatov808c43b2018-09-21 17:03:27 -0700585 while (cgrp && cgrp->level > ancestor_level)
586 cgrp = cgroup_parent(cgrp);
587 return cgrp;
Andrey Ignatov77236282018-08-12 10:49:27 -0700588}
589
590/**
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700591 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
592 * @task: the task to be tested
593 * @ancestor: possible ancestor of @task's cgroup
594 *
595 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
596 * It follows all the same rules as cgroup_is_descendant, and only applies
597 * to the default hierarchy.
598 */
599static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
600 struct cgroup *ancestor)
601{
602 struct css_set *cset = task_css_set(task);
603
604 return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
605}
606
Tejun Heo07bc3562014-02-13 06:58:39 -0500607/* no synchronization, the result can only be used as a hint */
Tejun Heo27bd4db2015-10-15 16:41:50 -0400608static inline bool cgroup_is_populated(struct cgroup *cgrp)
Tejun Heo07bc3562014-02-13 06:58:39 -0500609{
Tejun Heo454000a2017-05-15 09:34:02 -0400610 return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children +
611 cgrp->nr_populated_threaded_children;
Tejun Heo07bc3562014-02-13 06:58:39 -0500612}
613
Zefan Lif29374b2014-09-19 16:29:31 +0800614/* returns ino associated with a cgroup */
Tejun Heob1664922014-02-11 11:52:49 -0500615static inline ino_t cgroup_ino(struct cgroup *cgrp)
616{
Shaohua Lic53cd492017-07-12 11:49:50 -0700617 return cgrp->kn->id.ino;
Tejun Heob1664922014-02-11 11:52:49 -0500618}
619
Tejun Heob4168642014-05-13 12:16:21 -0400620/* cft/css accessors for cftype->write() operation */
621static inline struct cftype *of_cft(struct kernfs_open_file *of)
Tejun Heo7da11272013-12-05 12:28:04 -0500622{
Tejun Heo2bd59d42014-02-11 11:52:49 -0500623 return of->kn->priv;
Tejun Heo7da11272013-12-05 12:28:04 -0500624}
625
Tejun Heob4168642014-05-13 12:16:21 -0400626struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
627
628/* cft/css accessors for cftype->seq_*() operations */
629static inline struct cftype *seq_cft(struct seq_file *seq)
630{
631 return of_cft(seq->private);
632}
633
634static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
635{
636 return of_css(seq->private);
637}
Tejun Heo59f52962014-02-11 11:52:49 -0500638
Tejun Heoe61734c2014-02-12 09:29:50 -0500639/*
640 * Name / path handling functions. All are thin wrappers around the kernfs
641 * counterparts and can be called under any context.
642 */
643
644static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
645{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400646 return kernfs_name(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500647}
648
Tejun Heo4c737b42016-08-10 11:23:44 -0400649static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
Tejun Heoe61734c2014-02-12 09:29:50 -0500650{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400651 return kernfs_path(cgrp->kn, buf, buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500652}
653
654static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
655{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400656 pr_cont_kernfs_name(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500657}
658
659static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
660{
Tejun Heofdce6bf2014-03-19 10:23:54 -0400661 pr_cont_kernfs_path(cgrp->kn);
Tejun Heoe61734c2014-02-12 09:29:50 -0500662}
663
Johannes Weiner2ce71352018-10-26 15:06:31 -0700664static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
665{
666 return &cgrp->psi;
667}
668
Tejun Heo77f88792017-03-16 16:54:24 -0400669static inline void cgroup_init_kthreadd(void)
670{
671 /*
672 * kthreadd is inherited by all kthreads, keep it in the root so
673 * that the new kthreads are guaranteed to stay in the root until
674 * initialization is finished.
675 */
676 current->no_cgroup_migration = 1;
677}
678
679static inline void cgroup_kthread_ready(void)
680{
681 /*
682 * This kthread finished initialization. The creator should have
683 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
684 */
685 current->no_cgroup_migration = 0;
686}
687
Shaohua Li121508d2017-07-12 11:49:52 -0700688static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
689{
690 return &cgrp->kn->id;
691}
Shaohua Li69fd5c32017-07-12 11:49:55 -0700692
693void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
694 char *buf, size_t buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700695#else /* !CONFIG_CGROUPS */
696
Tejun Heof3ba5382015-01-06 12:02:46 -0500697struct cgroup_subsys_state;
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700698struct cgroup;
Tejun Heof3ba5382015-01-06 12:02:46 -0500699
Tejun Heoc326aa22015-05-13 16:24:16 -0400700static inline void css_put(struct cgroup_subsys_state *css) {}
701static inline int cgroup_attach_task_all(struct task_struct *from,
702 struct task_struct *t) { return 0; }
703static inline int cgroupstats_build(struct cgroupstats *stats,
704 struct dentry *dentry) { return -EINVAL; }
705
Paul Menageb4f48b62007-10-18 23:39:33 -0700706static inline void cgroup_fork(struct task_struct *p) {}
Oleg Nesterovb53202e2015-12-03 10:24:08 -0500707static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
708static inline void cgroup_cancel_fork(struct task_struct *p) {}
709static inline void cgroup_post_fork(struct task_struct *p) {}
Li Zefan1ec41832014-03-28 15:22:19 +0800710static inline void cgroup_exit(struct task_struct *p) {}
Oleg Nesterov51bee5a2019-01-28 17:00:13 +0100711static inline void cgroup_release(struct task_struct *p) {}
Tejun Heo2e91fa72015-10-15 16:41:53 -0400712static inline void cgroup_free(struct task_struct *p) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700713
Tejun Heoc326aa22015-05-13 16:24:16 -0400714static inline int cgroup_init_early(void) { return 0; }
715static inline int cgroup_init(void) { return 0; }
Tejun Heo77f88792017-03-16 16:54:24 -0400716static inline void cgroup_init_kthreadd(void) {}
717static inline void cgroup_kthread_ready(void) {}
Shaohua Li121508d2017-07-12 11:49:52 -0700718static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp)
719{
720 return NULL;
721}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200722
Johannes Weiner2ce71352018-10-26 15:06:31 -0700723static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
724{
725 return NULL;
726}
727
728static inline struct psi_group *cgroup_psi(struct cgroup *cgrp)
729{
730 return NULL;
731}
732
Sargun Dhillonaed704b2016-08-12 08:56:40 -0700733static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
734 struct cgroup *ancestor)
735{
736 return true;
737}
Shaohua Li69fd5c32017-07-12 11:49:55 -0700738
739static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
740 char *buf, size_t buflen) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700741#endif /* !CONFIG_CGROUPS */
742
Tejun Heo6162cef2018-04-26 14:29:05 -0700743#ifdef CONFIG_CGROUPS
744/*
745 * cgroup scalable recursive statistics.
746 */
747void cgroup_rstat_updated(struct cgroup *cgrp, int cpu);
748void cgroup_rstat_flush(struct cgroup *cgrp);
Tejun Heo0fa294f2018-04-26 14:29:05 -0700749void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp);
Tejun Heo6162cef2018-04-26 14:29:05 -0700750void cgroup_rstat_flush_hold(struct cgroup *cgrp);
751void cgroup_rstat_flush_release(void);
752
Tejun Heobd1060a2015-12-07 17:38:53 -0500753/*
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700754 * Basic resource stats.
755 */
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700756#ifdef CONFIG_CGROUP_CPUACCT
757void cpuacct_charge(struct task_struct *tsk, u64 cputime);
758void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);
759#else
760static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
761static inline void cpuacct_account_field(struct task_struct *tsk, int index,
762 u64 val) {}
763#endif
764
Tejun Heo041cd642017-09-25 08:12:05 -0700765void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec);
766void __cgroup_account_cputime_field(struct cgroup *cgrp,
767 enum cpu_usage_stat index, u64 delta_exec);
768
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700769static inline void cgroup_account_cputime(struct task_struct *task,
770 u64 delta_exec)
771{
Tejun Heo041cd642017-09-25 08:12:05 -0700772 struct cgroup *cgrp;
773
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700774 cpuacct_charge(task, delta_exec);
Tejun Heo041cd642017-09-25 08:12:05 -0700775
776 rcu_read_lock();
777 cgrp = task_dfl_cgroup(task);
778 if (cgroup_parent(cgrp))
779 __cgroup_account_cputime(cgrp, delta_exec);
780 rcu_read_unlock();
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700781}
782
783static inline void cgroup_account_cputime_field(struct task_struct *task,
784 enum cpu_usage_stat index,
785 u64 delta_exec)
786{
Tejun Heo041cd642017-09-25 08:12:05 -0700787 struct cgroup *cgrp;
788
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700789 cpuacct_account_field(task, index, delta_exec);
Tejun Heo041cd642017-09-25 08:12:05 -0700790
791 rcu_read_lock();
792 cgrp = task_dfl_cgroup(task);
793 if (cgroup_parent(cgrp))
794 __cgroup_account_cputime_field(cgrp, index, delta_exec);
795 rcu_read_unlock();
Tejun Heod2cc5ed2017-09-25 08:12:04 -0700796}
797
798#else /* CONFIG_CGROUPS */
799
800static inline void cgroup_account_cputime(struct task_struct *task,
801 u64 delta_exec) {}
802static inline void cgroup_account_cputime_field(struct task_struct *task,
803 enum cpu_usage_stat index,
804 u64 delta_exec) {}
805
806#endif /* CONFIG_CGROUPS */
807
808/*
Tejun Heobd1060a2015-12-07 17:38:53 -0500809 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
810 * definition in cgroup-defs.h.
811 */
812#ifdef CONFIG_SOCK_CGROUP_DATA
813
814#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
815extern spinlock_t cgroup_sk_update_lock;
816#endif
817
818void cgroup_sk_alloc_disable(void);
819void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
820void cgroup_sk_free(struct sock_cgroup_data *skcd);
821
822static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
823{
824#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
825 unsigned long v;
826
827 /*
828 * @skcd->val is 64bit but the following is safe on 32bit too as we
829 * just need the lower ulong to be written and read atomically.
830 */
831 v = READ_ONCE(skcd->val);
832
833 if (v & 1)
834 return &cgrp_dfl_root.cgrp;
835
836 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
837#else
838 return (struct cgroup *)(unsigned long)skcd->val;
839#endif
840}
841
842#else /* CONFIG_CGROUP_DATA */
843
844static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
845static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
846
847#endif /* CONFIG_CGROUP_DATA */
848
Aditya Kalia79a9082016-01-29 02:54:06 -0600849struct cgroup_namespace {
Elena Reshetova387ad962017-02-20 12:19:00 +0200850 refcount_t count;
Aditya Kalia79a9082016-01-29 02:54:06 -0600851 struct ns_common ns;
852 struct user_namespace *user_ns;
Eric W. Biedermand08311d2016-08-08 14:25:30 -0500853 struct ucounts *ucounts;
Aditya Kalia79a9082016-01-29 02:54:06 -0600854 struct css_set *root_cset;
855};
856
857extern struct cgroup_namespace init_cgroup_ns;
858
859#ifdef CONFIG_CGROUPS
860
861void free_cgroup_ns(struct cgroup_namespace *ns);
862
863struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
864 struct user_namespace *user_ns,
865 struct cgroup_namespace *old_ns);
866
Tejun Heo4c737b42016-08-10 11:23:44 -0400867int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
868 struct cgroup_namespace *ns);
Aditya Kalia79a9082016-01-29 02:54:06 -0600869
870#else /* !CONFIG_CGROUPS */
871
872static inline void free_cgroup_ns(struct cgroup_namespace *ns) { }
873static inline struct cgroup_namespace *
874copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns,
875 struct cgroup_namespace *old_ns)
876{
877 return old_ns;
878}
879
880#endif /* !CONFIG_CGROUPS */
881
882static inline void get_cgroup_ns(struct cgroup_namespace *ns)
883{
884 if (ns)
Elena Reshetova387ad962017-02-20 12:19:00 +0200885 refcount_inc(&ns->count);
Aditya Kalia79a9082016-01-29 02:54:06 -0600886}
887
888static inline void put_cgroup_ns(struct cgroup_namespace *ns)
889{
Elena Reshetova387ad962017-02-20 12:19:00 +0200890 if (ns && refcount_dec_and_test(&ns->count))
Aditya Kalia79a9082016-01-29 02:54:06 -0600891 free_cgroup_ns(ns);
892}
893
Roman Gushchin76f969e2019-04-19 10:03:04 -0700894#ifdef CONFIG_CGROUPS
895
896void cgroup_enter_frozen(void);
897void cgroup_leave_frozen(bool always_leave);
898void cgroup_update_frozen(struct cgroup *cgrp);
899void cgroup_freeze(struct cgroup *cgrp, bool freeze);
900void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src,
901 struct cgroup *dst);
Roman Gushchin96b9c592019-04-26 10:59:45 -0700902
Roman Gushchin76f969e2019-04-19 10:03:04 -0700903static inline bool cgroup_task_freeze(struct task_struct *task)
904{
905 bool ret;
906
907 if (task->flags & PF_KTHREAD)
908 return false;
909
910 rcu_read_lock();
911 ret = test_bit(CGRP_FREEZE, &task_dfl_cgroup(task)->flags);
912 rcu_read_unlock();
913
914 return ret;
915}
916
917static inline bool cgroup_task_frozen(struct task_struct *task)
918{
919 return task->frozen;
920}
921
922#else /* !CONFIG_CGROUPS */
923
924static inline void cgroup_enter_frozen(void) { }
925static inline void cgroup_leave_frozen(bool always_leave) { }
926static inline bool cgroup_task_freeze(struct task_struct *task)
927{
928 return false;
929}
930static inline bool cgroup_task_frozen(struct task_struct *task)
931{
932 return false;
933}
934
935#endif /* !CONFIG_CGROUPS */
936
Paul Menageddbcc7e2007-10-18 23:39:30 -0700937#endif /* _LINUX_CGROUP_H */