xref: /linux/include/linux/cgroup.h (revision 5b30afc20b3fea29b9beb83c6415c4ff06f774aa)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #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>
13 #include <linux/nodemask.h>
14 #include <linux/list.h>
15 #include <linux/rculist.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/fs.h>
18 #include <linux/seq_file.h>
19 #include <linux/kernfs.h>
20 #include <linux/jump_label.h>
21 #include <linux/types.h>
22 #include <linux/notifier.h>
23 #include <linux/ns_common.h>
24 #include <linux/nsproxy.h>
25 #include <linux/user_namespace.h>
26 #include <linux/refcount.h>
27 #include <linux/kernel_stat.h>
28 
29 #include <linux/cgroup-defs.h>
30 #include <linux/cgroup_namespace.h>
31 
32 struct kernel_clone_args;
33 
34 /*
35  * All weight knobs on the default hierarchy should use the following min,
36  * default and max values.  The default value is the logarithmic center of
37  * MIN and MAX and allows 100x to be expressed in both directions.
38  */
39 #define CGROUP_WEIGHT_MIN		1
40 #define CGROUP_WEIGHT_DFL		100
41 #define CGROUP_WEIGHT_MAX		10000
42 
43 #ifdef CONFIG_CGROUPS
44 
45 /*
46  * To avoid confusing the compiler (and generating warnings) with code
47  * that attempts to access what would be a 0-element array (i.e. sized
48  * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
49  * constant expression can be added.
50  */
51 #define CGROUP_HAS_SUBSYS_CONFIG	(CGROUP_SUBSYS_COUNT > 0)
52 
53 enum css_task_iter_flags {
54 	CSS_TASK_ITER_PROCS    = (1U << 0),  /* walk only threadgroup leaders */
55 	CSS_TASK_ITER_THREADED = (1U << 1),  /* walk all threaded css_sets in the domain */
56 	CSS_TASK_ITER_SKIPPED  = (1U << 16), /* internal flags */
57 };
58 
59 /* a css_task_iter should be treated as an opaque object */
60 struct css_task_iter {
61 	struct cgroup_subsys		*ss;
62 	unsigned int			flags;
63 
64 	struct list_head		*cset_pos;
65 	struct list_head		*cset_head;
66 
67 	struct list_head		*tcset_pos;
68 	struct list_head		*tcset_head;
69 
70 	struct list_head		*task_pos;
71 
72 	struct list_head		*cur_tasks_head;
73 	struct css_set			*cur_cset;
74 	struct css_set			*cur_dcset;
75 	struct task_struct		*cur_task;
76 	struct list_head		iters_node;	/* css_set->task_iters */
77 };
78 
79 enum cgroup_lifetime_events {
80 	CGROUP_LIFETIME_ONLINE,
81 	CGROUP_LIFETIME_OFFLINE,
82 };
83 
84 extern struct file_system_type cgroup_fs_type;
85 extern struct cgroup_root cgrp_dfl_root;
86 extern struct css_set init_css_set;
87 extern struct mutex cgroup_mutex;
88 extern spinlock_t css_set_lock;
89 extern struct blocking_notifier_head cgroup_lifetime_notifier;
90 
91 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
92 #include <linux/cgroup_subsys.h>
93 #undef SUBSYS
94 
95 #define SUBSYS(_x)								\
96 	extern struct static_key_true _x ## _cgrp_subsys_enabled_key;		\
97 	extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
98 #include <linux/cgroup_subsys.h>
99 #undef SUBSYS
100 
101 /**
102  * cgroup_subsys_enabled - fast test on whether a subsys is enabled
103  * @ss: subsystem in question
104  */
105 #define cgroup_subsys_enabled(ss)						\
106 	static_branch_likely(&ss ## _enabled_key)
107 
108 /**
109  * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
110  * @ss: subsystem in question
111  */
112 #define cgroup_subsys_on_dfl(ss)						\
113 	static_branch_likely(&ss ## _on_dfl_key)
114 
115 bool cgroup_on_dfl(const struct cgroup *cgrp);
116 
117 bool css_has_online_children(struct cgroup_subsys_state *css);
118 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
119 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup,
120 					 struct cgroup_subsys *ss);
121 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
122 					     struct cgroup_subsys *ss);
123 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
124 						       struct cgroup_subsys *ss);
125 
126 struct cgroup *cgroup_get_from_path(const char *path);
127 struct cgroup *cgroup_get_from_fd(int fd);
128 struct cgroup *cgroup_v1v2_get_from_fd(int fd);
129 
130 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
131 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
132 
133 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
134 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
135 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
136 int cgroup_rm_cftypes(struct cftype *cfts);
137 void cgroup_file_notify(struct cgroup_file *cfile);
138 void cgroup_file_show(struct cgroup_file *cfile, bool show);
139 
140 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
141 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
142 		     struct pid *pid, struct task_struct *tsk);
143 
144 void cgroup_fork(struct task_struct *p);
145 extern int cgroup_can_fork(struct task_struct *p,
146 			   struct kernel_clone_args *kargs);
147 extern void cgroup_cancel_fork(struct task_struct *p,
148 			       struct kernel_clone_args *kargs);
149 extern void cgroup_post_fork(struct task_struct *p,
150 			     struct kernel_clone_args *kargs);
151 void cgroup_task_exit(struct task_struct *p);
152 void cgroup_task_dead(struct task_struct *p);
153 void cgroup_task_release(struct task_struct *p);
154 void cgroup_task_free(struct task_struct *p);
155 
156 int cgroup_init_early(void);
157 int cgroup_init(void);
158 
159 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v);
160 
161 /*
162  * Iteration helpers and macros.
163  */
164 
165 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
166 					   struct cgroup_subsys_state *parent);
167 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
168 						    struct cgroup_subsys_state *css);
169 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
170 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
171 						     struct cgroup_subsys_state *css);
172 
173 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
174 					 struct cgroup_subsys_state **dst_cssp);
175 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
176 					struct cgroup_subsys_state **dst_cssp);
177 
178 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
179 			 struct css_task_iter *it);
180 struct task_struct *css_task_iter_next(struct css_task_iter *it);
181 void css_task_iter_end(struct css_task_iter *it);
182 
183 /**
184  * css_for_each_child - iterate through children of a css
185  * @pos: the css * to use as the loop cursor
186  * @parent: css whose children to walk
187  *
188  * Walk @parent's children.  Must be called under rcu_read_lock().
189  *
190  * If a subsystem synchronizes ->css_online() and the start of iteration, a
191  * css which finished ->css_online() is guaranteed to be visible in the
192  * future iterations and will stay visible until the last reference is put.
193  * A css which hasn't finished ->css_online() or already finished
194  * ->css_offline() may show up during traversal.  It's each subsystem's
195  * responsibility to synchronize against on/offlining.
196  *
197  * It is allowed to temporarily drop RCU read lock during iteration.  The
198  * caller is responsible for ensuring that @pos remains accessible until
199  * the start of the next iteration by, for example, bumping the css refcnt.
200  */
201 #define css_for_each_child(pos, parent)					\
202 	for ((pos) = css_next_child(NULL, (parent)); (pos);		\
203 	     (pos) = css_next_child((pos), (parent)))
204 
205 /**
206  * css_for_each_descendant_pre - pre-order walk of a css's descendants
207  * @pos: the css * to use as the loop cursor
208  * @root: css whose descendants to walk
209  *
210  * Walk @root's descendants.  @root is included in the iteration and the
211  * first node to be visited.  Must be called under rcu_read_lock().
212  *
213  * If a subsystem synchronizes ->css_online() and the start of iteration, a
214  * css which finished ->css_online() is guaranteed to be visible in the
215  * future iterations and will stay visible until the last reference is put.
216  * A css which hasn't finished ->css_online() or already finished
217  * ->css_offline() may show up during traversal.  It's each subsystem's
218  * responsibility to synchronize against on/offlining.
219  *
220  * For example, the following guarantees that a descendant can't escape
221  * state updates of its ancestors.
222  *
223  * my_online(@css)
224  * {
225  *	Lock @css's parent and @css;
226  *	Inherit state from the parent;
227  *	Unlock both.
228  * }
229  *
230  * my_update_state(@css)
231  * {
232  *	css_for_each_descendant_pre(@pos, @css) {
233  *		Lock @pos;
234  *		if (@pos == @css)
235  *			Update @css's state;
236  *		else
237  *			Verify @pos is alive and inherit state from its parent;
238  *		Unlock @pos;
239  *	}
240  * }
241  *
242  * As long as the inheriting step, including checking the parent state, is
243  * enclosed inside @pos locking, double-locking the parent isn't necessary
244  * while inheriting.  The state update to the parent is guaranteed to be
245  * visible by walking order and, as long as inheriting operations to the
246  * same @pos are atomic to each other, multiple updates racing each other
247  * still result in the correct state.  It's guaranateed that at least one
248  * inheritance happens for any css after the latest update to its parent.
249  *
250  * If checking parent's state requires locking the parent, each inheriting
251  * iteration should lock and unlock both @pos->parent and @pos.
252  *
253  * Alternatively, a subsystem may choose to use a single global lock to
254  * synchronize ->css_online() and ->css_offline() against tree-walking
255  * operations.
256  *
257  * It is allowed to temporarily drop RCU read lock during iteration.  The
258  * caller is responsible for ensuring that @pos remains accessible until
259  * the start of the next iteration by, for example, bumping the css refcnt.
260  */
261 #define css_for_each_descendant_pre(pos, css)				\
262 	for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);	\
263 	     (pos) = css_next_descendant_pre((pos), (css)))
264 
265 /**
266  * css_for_each_descendant_post - post-order walk of a css's descendants
267  * @pos: the css * to use as the loop cursor
268  * @css: css whose descendants to walk
269  *
270  * Similar to css_for_each_descendant_pre() but performs post-order
271  * traversal instead.  @root is included in the iteration and the last
272  * node to be visited.
273  *
274  * If a subsystem synchronizes ->css_online() and the start of iteration, a
275  * css which finished ->css_online() is guaranteed to be visible in the
276  * future iterations and will stay visible until the last reference is put.
277  * A css which hasn't finished ->css_online() or already finished
278  * ->css_offline() may show up during traversal.  It's each subsystem's
279  * responsibility to synchronize against on/offlining.
280  *
281  * Note that the walk visibility guarantee example described in pre-order
282  * walk doesn't apply the same to post-order walks.
283  */
284 #define css_for_each_descendant_post(pos, css)				\
285 	for ((pos) = css_next_descendant_post(NULL, (css)); (pos);	\
286 	     (pos) = css_next_descendant_post((pos), (css)))
287 
288 /* iterate over child cgrps, lock should be held throughout iteration */
289 #define cgroup_for_each_live_child(child, cgrp)				\
290 	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
291 		if (({ lockdep_assert_held(&cgroup_mutex);		\
292 		       cgroup_is_dead(child); }))			\
293 			;						\
294 		else
295 
296 /* walk live descendants in pre order */
297 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)		\
298 	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL))	\
299 		if (({ lockdep_assert_held(&cgroup_mutex);		\
300 		       (dsct) = (d_css)->cgroup;			\
301 		       cgroup_is_dead(dsct); }))			\
302 			;						\
303 		else
304 
305 /* walk live descendants in postorder */
306 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp)		\
307 	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL))	\
308 		if (({ lockdep_assert_held(&cgroup_mutex);		\
309 		       (dsct) = (d_css)->cgroup;			\
310 		       cgroup_is_dead(dsct); }))			\
311 			;						\
312 		else
313 
314 /**
315  * cgroup_taskset_for_each - iterate cgroup_taskset
316  * @task: the loop cursor
317  * @dst_css: the destination css
318  * @tset: taskset to iterate
319  *
320  * @tset may contain multiple tasks and they may belong to multiple
321  * processes.
322  *
323  * On the v2 hierarchy, there may be tasks from multiple processes and they
324  * may not share the source or destination csses.
325  *
326  * On traditional hierarchies, when there are multiple tasks in @tset, if a
327  * task of a process is in @tset, all tasks of the process are in @tset.
328  * Also, all are guaranteed to share the same source and destination csses.
329  *
330  * Iteration is not in any specific order.
331  */
332 #define cgroup_taskset_for_each(task, dst_css, tset)			\
333 	for ((task) = cgroup_taskset_first((tset), &(dst_css));		\
334 	     (task);							\
335 	     (task) = cgroup_taskset_next((tset), &(dst_css)))
336 
337 /**
338  * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
339  * @leader: the loop cursor
340  * @dst_css: the destination css
341  * @tset: taskset to iterate
342  *
343  * Iterate threadgroup leaders of @tset.  For single-task migrations, @tset
344  * may not contain any.
345  */
346 #define cgroup_taskset_for_each_leader(leader, dst_css, tset)		\
347 	for ((leader) = cgroup_taskset_first((tset), &(dst_css));	\
348 	     (leader);							\
349 	     (leader) = cgroup_taskset_next((tset), &(dst_css)))	\
350 		if ((leader) != (leader)->group_leader)			\
351 			;						\
352 		else
353 
354 /*
355  * Inline functions.
356  */
357 
358 #ifdef CONFIG_DEBUG_CGROUP_REF
359 void css_get(struct cgroup_subsys_state *css);
360 void css_get_many(struct cgroup_subsys_state *css, unsigned int n);
361 bool css_tryget(struct cgroup_subsys_state *css);
362 bool css_tryget_online(struct cgroup_subsys_state *css);
363 void css_put(struct cgroup_subsys_state *css);
364 void css_put_many(struct cgroup_subsys_state *css, unsigned int n);
365 #else
366 #define CGROUP_REF_FN_ATTRS	static inline
367 #define CGROUP_REF_EXPORT(fn)
368 #include <linux/cgroup_refcnt.h>
369 #endif
370 
371 static inline u64 cgroup_id(const struct cgroup *cgrp)
372 {
373 	return cgrp->kn->id;
374 }
375 
376 /**
377  * cgroup_css - obtain a cgroup's css for the specified subsystem
378  * @cgrp: the cgroup of interest
379  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
380  *
381  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
382  * function must be called either under cgroup_mutex or rcu_read_lock() and
383  * the caller is responsible for pinning the returned css if it wants to
384  * keep accessing it outside the said locks.  This function may return
385  * %NULL if @cgrp doesn't have @subsys_id enabled.
386  */
387 static inline struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
388 						     struct cgroup_subsys *ss)
389 {
390 	if (CGROUP_HAS_SUBSYS_CONFIG && ss)
391 		return rcu_dereference_check(cgrp->subsys[ss->id],
392 					lockdep_is_held(&cgroup_mutex));
393 	else
394 		return &cgrp->self;
395 }
396 
397 /**
398  * css_is_dying - test whether the specified css is dying
399  * @css: target css
400  *
401  * Test whether @css is in the process of offlining or already offline.  In
402  * most cases, ->css_online() and ->css_offline() callbacks should be
403  * enough; however, the actual offline operations are RCU delayed and this
404  * test returns %true also when @css is scheduled to be offlined.
405  *
406  * This is useful, for example, when the use case requires synchronous
407  * behavior with respect to cgroup removal.  cgroup removal schedules css
408  * offlining but the css can seem alive while the operation is being
409  * delayed.  If the delay affects user visible semantics, this test can be
410  * used to resolve the situation.
411  */
412 static inline bool css_is_dying(struct cgroup_subsys_state *css)
413 {
414 	return css->flags & CSS_DYING;
415 }
416 
417 static inline bool css_is_online(struct cgroup_subsys_state *css)
418 {
419 	return css->flags & CSS_ONLINE;
420 }
421 
422 static inline bool css_is_self(struct cgroup_subsys_state *css)
423 {
424 	if (css == &css->cgroup->self) {
425 		/* cgroup::self should not have subsystem association */
426 		WARN_ON(css->ss != NULL);
427 		return true;
428 	}
429 
430 	return false;
431 }
432 
433 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
434 {
435 	return !(cgrp->self.flags & CSS_ONLINE);
436 }
437 
438 static inline void cgroup_get(struct cgroup *cgrp)
439 {
440 	css_get(&cgrp->self);
441 }
442 
443 static inline bool cgroup_tryget(struct cgroup *cgrp)
444 {
445 	return css_tryget(&cgrp->self);
446 }
447 
448 static inline void cgroup_put(struct cgroup *cgrp)
449 {
450 	css_put(&cgrp->self);
451 }
452 
453 static inline void cgroup_lock(void)
454 {
455 	mutex_lock(&cgroup_mutex);
456 }
457 
458 static inline void cgroup_unlock(void)
459 {
460 	mutex_unlock(&cgroup_mutex);
461 }
462 
463 /**
464  * task_css_set_check - obtain a task's css_set with extra access conditions
465  * @task: the task to obtain css_set for
466  * @__c: extra condition expression to be passed to rcu_dereference_check()
467  *
468  * A task's css_set is RCU protected, initialized and exited while holding
469  * task_lock(), and can only be modified while holding both cgroup_mutex
470  * and task_lock() while the task is alive.  This macro verifies that the
471  * caller is inside proper critical section and returns @task's css_set.
472  *
473  * The caller can also specify additional allowed conditions via @__c, such
474  * as locks used during the cgroup_subsys::attach() methods.
475  */
476 #ifdef CONFIG_PROVE_RCU
477 #define task_css_set_check(task, __c)					\
478 	rcu_dereference_check((task)->cgroups,				\
479 		rcu_read_lock_sched_held() ||				\
480 		lockdep_is_held(&cgroup_mutex) ||			\
481 		lockdep_is_held(&css_set_lock) ||			\
482 		((task)->flags & PF_EXITING) || (__c))
483 #else
484 #define task_css_set_check(task, __c)					\
485 	rcu_dereference((task)->cgroups)
486 #endif
487 
488 /**
489  * task_css_check - obtain css for (task, subsys) w/ extra access conds
490  * @task: the target task
491  * @subsys_id: the target subsystem ID
492  * @__c: extra condition expression to be passed to rcu_dereference_check()
493  *
494  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
495  * synchronization rules are the same as task_css_set_check().
496  */
497 #define task_css_check(task, subsys_id, __c)				\
498 	task_css_set_check((task), (__c))->subsys[(subsys_id)]
499 
500 /**
501  * task_css_set - obtain a task's css_set
502  * @task: the task to obtain css_set for
503  *
504  * See task_css_set_check().
505  */
506 static inline struct css_set *task_css_set(struct task_struct *task)
507 {
508 	return task_css_set_check(task, false);
509 }
510 
511 /**
512  * task_css - obtain css for (task, subsys)
513  * @task: the target task
514  * @subsys_id: the target subsystem ID
515  *
516  * See task_css_check().
517  */
518 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
519 						   int subsys_id)
520 {
521 	return task_css_check(task, subsys_id, false);
522 }
523 
524 /**
525  * task_get_css - find and get the css for (task, subsys)
526  * @task: the target task
527  * @subsys_id: the target subsystem ID
528  *
529  * Find the css for the (@task, @subsys_id) combination, increment a
530  * reference on and return it.  This function is guaranteed to return a
531  * valid css.  The returned css may already have been offlined.
532  */
533 static inline struct cgroup_subsys_state *
534 task_get_css(struct task_struct *task, int subsys_id)
535 {
536 	struct cgroup_subsys_state *css;
537 
538 	rcu_read_lock();
539 	while (true) {
540 		css = task_css(task, subsys_id);
541 		/*
542 		 * Can't use css_tryget_online() here.  A task which has
543 		 * PF_EXITING set may stay associated with an offline css.
544 		 * If such task calls this function, css_tryget_online()
545 		 * will keep failing.
546 		 */
547 		if (likely(css_tryget(css)))
548 			break;
549 		cpu_relax();
550 	}
551 	rcu_read_unlock();
552 	return css;
553 }
554 
555 /**
556  * task_css_is_root - test whether a task belongs to the root css
557  * @task: the target task
558  * @subsys_id: the target subsystem ID
559  *
560  * Test whether @task belongs to the root css on the specified subsystem.
561  * May be invoked in any context.
562  */
563 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
564 {
565 	return task_css_check(task, subsys_id, true) ==
566 		init_css_set.subsys[subsys_id];
567 }
568 
569 static inline struct cgroup *task_cgroup(struct task_struct *task,
570 					 int subsys_id)
571 {
572 	return task_css(task, subsys_id)->cgroup;
573 }
574 
575 static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
576 {
577 	return task_css_set(task)->dfl_cgrp;
578 }
579 
580 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
581 {
582 	struct cgroup_subsys_state *parent_css = cgrp->self.parent;
583 
584 	if (parent_css)
585 		return container_of(parent_css, struct cgroup, self);
586 	return NULL;
587 }
588 
589 /**
590  * cgroup_is_descendant - test ancestry
591  * @cgrp: the cgroup to be tested
592  * @ancestor: possible ancestor of @cgrp
593  *
594  * Test whether @cgrp is a descendant of @ancestor.  It also returns %true
595  * if @cgrp == @ancestor.  This function is safe to call as long as @cgrp
596  * and @ancestor are accessible.
597  */
598 static inline bool cgroup_is_descendant(struct cgroup *cgrp,
599 					struct cgroup *ancestor)
600 {
601 	if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
602 		return false;
603 	return cgrp->ancestors[ancestor->level] == ancestor;
604 }
605 
606 /**
607  * cgroup_ancestor - find ancestor of cgroup
608  * @cgrp: cgroup to find ancestor of
609  * @ancestor_level: level of ancestor to find starting from root
610  *
611  * Find ancestor of cgroup at specified level starting from root if it exists
612  * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at
613  * @ancestor_level.
614  *
615  * This function is safe to call as long as @cgrp is accessible.
616  */
617 static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
618 					     int ancestor_level)
619 {
620 	if (ancestor_level < 0 || ancestor_level > cgrp->level)
621 		return NULL;
622 	return cgrp->ancestors[ancestor_level];
623 }
624 
625 /**
626  * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
627  * @task: the task to be tested
628  * @ancestor: possible ancestor of @task's cgroup
629  *
630  * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
631  * It follows all the same rules as cgroup_is_descendant, and only applies
632  * to the default hierarchy.
633  */
634 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
635 					       struct cgroup *ancestor)
636 {
637 	struct css_set *cset = task_css_set(task);
638 
639 	return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
640 }
641 
642 /* no synchronization, the result can only be used as a hint */
643 static inline bool cgroup_is_populated(struct cgroup *cgrp)
644 {
645 	return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children +
646 		cgrp->nr_populated_threaded_children;
647 }
648 
649 /* returns ino associated with a cgroup */
650 static inline ino_t cgroup_ino(struct cgroup *cgrp)
651 {
652 	return kernfs_ino(cgrp->kn);
653 }
654 
655 /* cft/css accessors for cftype->write() operation */
656 static inline struct cftype *of_cft(struct kernfs_open_file *of)
657 {
658 	return of->kn->priv;
659 }
660 
661 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
662 
663 /* cft/css accessors for cftype->seq_*() operations */
664 static inline struct cftype *seq_cft(struct seq_file *seq)
665 {
666 	return of_cft(seq->private);
667 }
668 
669 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
670 {
671 	return of_css(seq->private);
672 }
673 
674 /*
675  * Name / path handling functions.  All are thin wrappers around the kernfs
676  * counterparts and can be called under any context.
677  */
678 
679 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
680 {
681 	return kernfs_name(cgrp->kn, buf, buflen);
682 }
683 
684 static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
685 {
686 	return kernfs_path(cgrp->kn, buf, buflen);
687 }
688 
689 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
690 {
691 	pr_cont_kernfs_name(cgrp->kn);
692 }
693 
694 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
695 {
696 	pr_cont_kernfs_path(cgrp->kn);
697 }
698 
699 bool cgroup_psi_enabled(void);
700 
701 static inline void cgroup_init_kthreadd(void)
702 {
703 	/*
704 	 * kthreadd is inherited by all kthreads, keep it in the root so
705 	 * that the new kthreads are guaranteed to stay in the root until
706 	 * initialization is finished.
707 	 */
708 	current->no_cgroup_migration = 1;
709 }
710 
711 static inline void cgroup_kthread_ready(void)
712 {
713 	/*
714 	 * This kthread finished initialization.  The creator should have
715 	 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
716 	 */
717 	current->no_cgroup_migration = 0;
718 }
719 
720 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen);
721 struct cgroup *__cgroup_get_from_id(u64 id);
722 struct cgroup *cgroup_get_from_id(u64 id);
723 #else /* !CONFIG_CGROUPS */
724 
725 struct cgroup_subsys_state;
726 struct cgroup;
727 
728 static inline u64 cgroup_id(const struct cgroup *cgrp) { return 1; }
729 static inline void css_get(struct cgroup_subsys_state *css) {}
730 static inline void css_put(struct cgroup_subsys_state *css) {}
731 static inline void cgroup_lock(void) {}
732 static inline void cgroup_unlock(void) {}
733 static inline int cgroup_attach_task_all(struct task_struct *from,
734 					 struct task_struct *t) { return 0; }
735 static inline int cgroupstats_build(struct cgroupstats *stats,
736 				    struct dentry *dentry) { return -EINVAL; }
737 
738 static inline void cgroup_fork(struct task_struct *p) {}
739 static inline int cgroup_can_fork(struct task_struct *p,
740 				  struct kernel_clone_args *kargs) { return 0; }
741 static inline void cgroup_cancel_fork(struct task_struct *p,
742 				      struct kernel_clone_args *kargs) {}
743 static inline void cgroup_post_fork(struct task_struct *p,
744 				    struct kernel_clone_args *kargs) {}
745 static inline void cgroup_task_exit(struct task_struct *p) {}
746 static inline void cgroup_task_dead(struct task_struct *p) {}
747 static inline void cgroup_task_release(struct task_struct *p) {}
748 static inline void cgroup_task_free(struct task_struct *p) {}
749 
750 static inline int cgroup_init_early(void) { return 0; }
751 static inline int cgroup_init(void) { return 0; }
752 static inline void cgroup_init_kthreadd(void) {}
753 static inline void cgroup_kthread_ready(void) {}
754 
755 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
756 {
757 	return NULL;
758 }
759 
760 static inline bool cgroup_psi_enabled(void)
761 {
762 	return false;
763 }
764 
765 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
766 					       struct cgroup *ancestor)
767 {
768 	return true;
769 }
770 
771 static inline void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
772 {}
773 #endif /* !CONFIG_CGROUPS */
774 
775 #ifdef CONFIG_CGROUPS
776 /*
777  * cgroup scalable recursive statistics.
778  */
779 void css_rstat_updated(struct cgroup_subsys_state *css, int cpu);
780 void css_rstat_flush(struct cgroup_subsys_state *css);
781 
782 /*
783  * Basic resource stats.
784  */
785 #ifdef CONFIG_CGROUP_CPUACCT
786 void cpuacct_charge(struct task_struct *tsk, u64 cputime);
787 void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);
788 #else
789 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
790 static inline void cpuacct_account_field(struct task_struct *tsk, int index,
791 					 u64 val) {}
792 #endif
793 
794 void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec);
795 void __cgroup_account_cputime_field(struct cgroup *cgrp,
796 				    enum cpu_usage_stat index, u64 delta_exec);
797 
798 static inline void cgroup_account_cputime(struct task_struct *task,
799 					  u64 delta_exec)
800 {
801 	struct cgroup *cgrp;
802 
803 	cpuacct_charge(task, delta_exec);
804 
805 	cgrp = task_dfl_cgroup(task);
806 	if (cgroup_parent(cgrp))
807 		__cgroup_account_cputime(cgrp, delta_exec);
808 }
809 
810 static inline void cgroup_account_cputime_field(struct task_struct *task,
811 						enum cpu_usage_stat index,
812 						u64 delta_exec)
813 {
814 	struct cgroup *cgrp;
815 
816 	cpuacct_account_field(task, index, delta_exec);
817 
818 	cgrp = task_dfl_cgroup(task);
819 	if (cgroup_parent(cgrp))
820 		__cgroup_account_cputime_field(cgrp, index, delta_exec);
821 }
822 
823 #else	/* CONFIG_CGROUPS */
824 
825 static inline void cgroup_account_cputime(struct task_struct *task,
826 					  u64 delta_exec) {}
827 static inline void cgroup_account_cputime_field(struct task_struct *task,
828 						enum cpu_usage_stat index,
829 						u64 delta_exec) {}
830 
831 #endif	/* CONFIG_CGROUPS */
832 
833 /*
834  * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
835  * definition in cgroup-defs.h.
836  */
837 #ifdef CONFIG_SOCK_CGROUP_DATA
838 
839 void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
840 void cgroup_sk_clone(struct sock_cgroup_data *skcd);
841 void cgroup_sk_free(struct sock_cgroup_data *skcd);
842 
843 static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
844 {
845 	return skcd->cgroup;
846 }
847 
848 #else	/* CONFIG_CGROUP_DATA */
849 
850 static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
851 static inline void cgroup_sk_clone(struct sock_cgroup_data *skcd) {}
852 static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
853 
854 #endif	/* CONFIG_CGROUP_DATA */
855 
856 #ifdef CONFIG_CGROUPS
857 
858 void cgroup_enter_frozen(void);
859 void cgroup_leave_frozen(bool always_leave);
860 void cgroup_update_frozen(struct cgroup *cgrp);
861 void cgroup_freeze(struct cgroup *cgrp, bool freeze);
862 void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src,
863 				 struct cgroup *dst);
864 
865 static inline bool cgroup_task_frozen(struct task_struct *task)
866 {
867 	return task->frozen;
868 }
869 
870 #else /* !CONFIG_CGROUPS */
871 
872 static inline void cgroup_enter_frozen(void) { }
873 static inline void cgroup_leave_frozen(bool always_leave) { }
874 static inline bool cgroup_task_frozen(struct task_struct *task)
875 {
876 	return false;
877 }
878 
879 #endif /* !CONFIG_CGROUPS */
880 
881 #ifdef CONFIG_CGROUP_BPF
882 static inline void cgroup_bpf_get(struct cgroup *cgrp)
883 {
884 	percpu_ref_get(&cgrp->bpf.refcnt);
885 }
886 
887 static inline void cgroup_bpf_put(struct cgroup *cgrp)
888 {
889 	percpu_ref_put(&cgrp->bpf.refcnt);
890 }
891 
892 #else /* CONFIG_CGROUP_BPF */
893 
894 static inline void cgroup_bpf_get(struct cgroup *cgrp) {}
895 static inline void cgroup_bpf_put(struct cgroup *cgrp) {}
896 
897 #endif /* CONFIG_CGROUP_BPF */
898 
899 struct cgroup *task_get_cgroup1(struct task_struct *tsk, int hierarchy_id);
900 
901 struct cgroup_of_peak *of_peak(struct kernfs_open_file *of);
902 
903 #endif /* _LINUX_CGROUP_H */
904