xref: /linux/include/linux/cgroup.h (revision 83476cc97bc635a3ff502bd194c79bfb1f1ae050)
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_WITH_DEAD = (1U << 2),  /* include exiting tasks */
57 	CSS_TASK_ITER_SKIPPED  = (1U << 16), /* internal flags */
58 };
59 
60 /* a css_task_iter should be treated as an opaque object */
61 struct css_task_iter {
62 	struct cgroup_subsys		*ss;
63 	unsigned int			flags;
64 
65 	struct list_head		*cset_pos;
66 	struct list_head		*cset_head;
67 
68 	struct list_head		*tcset_pos;
69 	struct list_head		*tcset_head;
70 
71 	struct list_head		*task_pos;
72 
73 	struct list_head		*cur_tasks_head;
74 	struct css_set			*cur_cset;
75 	struct css_set			*cur_dcset;
76 	struct task_struct		*cur_task;
77 	struct list_head		iters_node;	/* css_set->task_iters */
78 };
79 
80 enum cgroup_lifetime_events {
81 	CGROUP_LIFETIME_ONLINE,
82 	CGROUP_LIFETIME_OFFLINE,
83 };
84 
85 extern struct file_system_type cgroup_fs_type;
86 extern struct cgroup_root cgrp_dfl_root;
87 extern struct css_set init_css_set;
88 extern struct mutex cgroup_mutex;
89 extern spinlock_t css_set_lock;
90 extern struct blocking_notifier_head cgroup_lifetime_notifier;
91 
92 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
93 #include <linux/cgroup_subsys.h>
94 #undef SUBSYS
95 
96 #define SUBSYS(_x)								\
97 	extern struct static_key_true _x ## _cgrp_subsys_enabled_key;		\
98 	extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
99 #include <linux/cgroup_subsys.h>
100 #undef SUBSYS
101 
102 /**
103  * cgroup_subsys_enabled - fast test on whether a subsys is enabled
104  * @ss: subsystem in question
105  */
106 #define cgroup_subsys_enabled(ss)						\
107 	static_branch_likely(&ss ## _enabled_key)
108 
109 /**
110  * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
111  * @ss: subsystem in question
112  */
113 #define cgroup_subsys_on_dfl(ss)						\
114 	static_branch_likely(&ss ## _on_dfl_key)
115 
116 bool cgroup_on_dfl(const struct cgroup *cgrp);
117 
118 bool css_has_online_children(struct cgroup_subsys_state *css);
119 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
120 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup,
121 					 struct cgroup_subsys *ss);
122 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
123 					     struct cgroup_subsys *ss);
124 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
125 						       struct cgroup_subsys *ss);
126 
127 struct cgroup *cgroup_get_from_path(const char *path);
128 struct cgroup *cgroup_get_from_fd(int fd);
129 struct cgroup *cgroup_v1v2_get_from_fd(int fd);
130 
131 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
132 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
133 
134 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
135 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
136 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
137 int cgroup_rm_cftypes(struct cftype *cfts);
138 void cgroup_file_notify(struct cgroup_file *cfile);
139 void cgroup_file_show(struct cgroup_file *cfile, bool show);
140 
141 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
142 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
143 		     struct pid *pid, struct task_struct *tsk);
144 
145 void cgroup_fork(struct task_struct *p);
146 extern int cgroup_can_fork(struct task_struct *p,
147 			   struct kernel_clone_args *kargs);
148 extern void cgroup_cancel_fork(struct task_struct *p,
149 			       struct kernel_clone_args *kargs);
150 extern void cgroup_post_fork(struct task_struct *p,
151 			     struct kernel_clone_args *kargs);
152 void cgroup_task_exit(struct task_struct *p);
153 void cgroup_task_dead(struct task_struct *p);
154 void cgroup_task_release(struct task_struct *p);
155 void cgroup_task_free(struct task_struct *p);
156 
157 int cgroup_init_early(void);
158 int cgroup_init(void);
159 
160 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v);
161 
162 /*
163  * Iteration helpers and macros.
164  */
165 
166 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
167 					   struct cgroup_subsys_state *parent);
168 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
169 						    struct cgroup_subsys_state *css);
170 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
171 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
172 						     struct cgroup_subsys_state *css);
173 
174 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
175 					 struct cgroup_subsys_state **dst_cssp);
176 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
177 					struct cgroup_subsys_state **dst_cssp);
178 
179 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
180 			 struct css_task_iter *it);
181 struct task_struct *css_task_iter_next(struct css_task_iter *it);
182 void css_task_iter_end(struct css_task_iter *it);
183 
184 /**
185  * css_for_each_child - iterate through children of a css
186  * @pos: the css * to use as the loop cursor
187  * @parent: css whose children to walk
188  *
189  * Walk @parent's children.  Must be called under rcu_read_lock().
190  *
191  * If a subsystem synchronizes ->css_online() and the start of iteration, a
192  * css which finished ->css_online() is guaranteed to be visible in the
193  * future iterations and will stay visible until the last reference is put.
194  * A css which hasn't finished ->css_online() or already finished
195  * ->css_offline() may show up during traversal.  It's each subsystem's
196  * responsibility to synchronize against on/offlining.
197  *
198  * It is allowed to temporarily drop RCU read lock during iteration.  The
199  * caller is responsible for ensuring that @pos remains accessible until
200  * the start of the next iteration by, for example, bumping the css refcnt.
201  */
202 #define css_for_each_child(pos, parent)					\
203 	for ((pos) = css_next_child(NULL, (parent)); (pos);		\
204 	     (pos) = css_next_child((pos), (parent)))
205 
206 /**
207  * css_for_each_descendant_pre - pre-order walk of a css's descendants
208  * @pos: the css * to use as the loop cursor
209  * @root: css whose descendants to walk
210  *
211  * Walk @root's descendants.  @root is included in the iteration and the
212  * first node to be visited.  Must be called under rcu_read_lock().
213  *
214  * If a subsystem synchronizes ->css_online() and the start of iteration, a
215  * css which finished ->css_online() is guaranteed to be visible in the
216  * future iterations and will stay visible until the last reference is put.
217  * A css which hasn't finished ->css_online() or already finished
218  * ->css_offline() may show up during traversal.  It's each subsystem's
219  * responsibility to synchronize against on/offlining.
220  *
221  * For example, the following guarantees that a descendant can't escape
222  * state updates of its ancestors.
223  *
224  * my_online(@css)
225  * {
226  *	Lock @css's parent and @css;
227  *	Inherit state from the parent;
228  *	Unlock both.
229  * }
230  *
231  * my_update_state(@css)
232  * {
233  *	css_for_each_descendant_pre(@pos, @css) {
234  *		Lock @pos;
235  *		if (@pos == @css)
236  *			Update @css's state;
237  *		else
238  *			Verify @pos is alive and inherit state from its parent;
239  *		Unlock @pos;
240  *	}
241  * }
242  *
243  * As long as the inheriting step, including checking the parent state, is
244  * enclosed inside @pos locking, double-locking the parent isn't necessary
245  * while inheriting.  The state update to the parent is guaranteed to be
246  * visible by walking order and, as long as inheriting operations to the
247  * same @pos are atomic to each other, multiple updates racing each other
248  * still result in the correct state.  It's guaranateed that at least one
249  * inheritance happens for any css after the latest update to its parent.
250  *
251  * If checking parent's state requires locking the parent, each inheriting
252  * iteration should lock and unlock both @pos->parent and @pos.
253  *
254  * Alternatively, a subsystem may choose to use a single global lock to
255  * synchronize ->css_online() and ->css_offline() against tree-walking
256  * operations.
257  *
258  * It is allowed to temporarily drop RCU read lock during iteration.  The
259  * caller is responsible for ensuring that @pos remains accessible until
260  * the start of the next iteration by, for example, bumping the css refcnt.
261  */
262 #define css_for_each_descendant_pre(pos, css)				\
263 	for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);	\
264 	     (pos) = css_next_descendant_pre((pos), (css)))
265 
266 /**
267  * css_for_each_descendant_post - post-order walk of a css's descendants
268  * @pos: the css * to use as the loop cursor
269  * @css: css whose descendants to walk
270  *
271  * Similar to css_for_each_descendant_pre() but performs post-order
272  * traversal instead.  @root is included in the iteration and the last
273  * node to be visited.
274  *
275  * If a subsystem synchronizes ->css_online() and the start of iteration, a
276  * css which finished ->css_online() is guaranteed to be visible in the
277  * future iterations and will stay visible until the last reference is put.
278  * A css which hasn't finished ->css_online() or already finished
279  * ->css_offline() may show up during traversal.  It's each subsystem's
280  * responsibility to synchronize against on/offlining.
281  *
282  * Note that the walk visibility guarantee example described in pre-order
283  * walk doesn't apply the same to post-order walks.
284  */
285 #define css_for_each_descendant_post(pos, css)				\
286 	for ((pos) = css_next_descendant_post(NULL, (css)); (pos);	\
287 	     (pos) = css_next_descendant_post((pos), (css)))
288 
289 /* iterate over child cgrps, lock should be held throughout iteration */
290 #define cgroup_for_each_live_child(child, cgrp)				\
291 	list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
292 		if (({ lockdep_assert_held(&cgroup_mutex);		\
293 		       cgroup_is_dead(child); }))			\
294 			;						\
295 		else
296 
297 /* walk live descendants in pre order */
298 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)		\
299 	css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL))	\
300 		if (({ lockdep_assert_held(&cgroup_mutex);		\
301 		       (dsct) = (d_css)->cgroup;			\
302 		       cgroup_is_dead(dsct); }))			\
303 			;						\
304 		else
305 
306 /* walk live descendants in postorder */
307 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp)		\
308 	css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL))	\
309 		if (({ lockdep_assert_held(&cgroup_mutex);		\
310 		       (dsct) = (d_css)->cgroup;			\
311 		       cgroup_is_dead(dsct); }))			\
312 			;						\
313 		else
314 
315 /**
316  * cgroup_taskset_for_each - iterate cgroup_taskset
317  * @task: the loop cursor
318  * @dst_css: the destination css
319  * @tset: taskset to iterate
320  *
321  * @tset may contain multiple tasks and they may belong to multiple
322  * processes.
323  *
324  * On the v2 hierarchy, there may be tasks from multiple processes and they
325  * may not share the source or destination csses.
326  *
327  * On traditional hierarchies, when there are multiple tasks in @tset, if a
328  * task of a process is in @tset, all tasks of the process are in @tset.
329  * Also, all are guaranteed to share the same source and destination csses.
330  *
331  * Iteration is not in any specific order.
332  */
333 #define cgroup_taskset_for_each(task, dst_css, tset)			\
334 	for ((task) = cgroup_taskset_first((tset), &(dst_css));		\
335 	     (task);							\
336 	     (task) = cgroup_taskset_next((tset), &(dst_css)))
337 
338 /**
339  * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
340  * @leader: the loop cursor
341  * @dst_css: the destination css
342  * @tset: taskset to iterate
343  *
344  * Iterate threadgroup leaders of @tset.  For single-task migrations, @tset
345  * may not contain any.
346  */
347 #define cgroup_taskset_for_each_leader(leader, dst_css, tset)		\
348 	for ((leader) = cgroup_taskset_first((tset), &(dst_css));	\
349 	     (leader);							\
350 	     (leader) = cgroup_taskset_next((tset), &(dst_css)))	\
351 		if ((leader) != (leader)->group_leader)			\
352 			;						\
353 		else
354 
355 /*
356  * Inline functions.
357  */
358 
359 #ifdef CONFIG_DEBUG_CGROUP_REF
360 void css_get(struct cgroup_subsys_state *css);
361 void css_get_many(struct cgroup_subsys_state *css, unsigned int n);
362 bool css_tryget(struct cgroup_subsys_state *css);
363 bool css_tryget_online(struct cgroup_subsys_state *css);
364 void css_put(struct cgroup_subsys_state *css);
365 void css_put_many(struct cgroup_subsys_state *css, unsigned int n);
366 #else
367 #define CGROUP_REF_FN_ATTRS	static inline
368 #define CGROUP_REF_EXPORT(fn)
369 #include <linux/cgroup_refcnt.h>
370 #endif
371 
372 static inline u64 cgroup_id(const struct cgroup *cgrp)
373 {
374 	return cgrp->kn->id;
375 }
376 
377 /**
378  * cgroup_css - obtain a cgroup's css for the specified subsystem
379  * @cgrp: the cgroup of interest
380  * @ss: the subsystem of interest (%NULL returns @cgrp->self)
381  *
382  * Return @cgrp's css (cgroup_subsys_state) associated with @ss.  This
383  * function must be called either under cgroup_mutex or rcu_read_lock() and
384  * the caller is responsible for pinning the returned css if it wants to
385  * keep accessing it outside the said locks.  This function may return
386  * %NULL if @cgrp doesn't have @subsys_id enabled.
387  */
388 static inline struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
389 						     struct cgroup_subsys *ss)
390 {
391 	if (CGROUP_HAS_SUBSYS_CONFIG && ss)
392 		return rcu_dereference_check(cgrp->subsys[ss->id],
393 					lockdep_is_held(&cgroup_mutex));
394 	else
395 		return &cgrp->self;
396 }
397 
398 /**
399  * css_is_dying - test whether the specified css is dying
400  * @css: target css
401  *
402  * Test whether @css is in the process of offlining or already offline.  In
403  * most cases, ->css_online() and ->css_offline() callbacks should be
404  * enough; however, the actual offline operations are RCU delayed and this
405  * test returns %true also when @css is scheduled to be offlined.
406  *
407  * This is useful, for example, when the use case requires synchronous
408  * behavior with respect to cgroup removal.  cgroup removal schedules css
409  * offlining but the css can seem alive while the operation is being
410  * delayed.  If the delay affects user visible semantics, this test can be
411  * used to resolve the situation.
412  */
413 static inline bool css_is_dying(struct cgroup_subsys_state *css)
414 {
415 	return css->flags & CSS_DYING;
416 }
417 
418 static inline bool css_is_online(struct cgroup_subsys_state *css)
419 {
420 	return css->flags & CSS_ONLINE;
421 }
422 
423 static inline bool css_is_self(struct cgroup_subsys_state *css)
424 {
425 	if (css == &css->cgroup->self) {
426 		/* cgroup::self should not have subsystem association */
427 		WARN_ON(css->ss != NULL);
428 		return true;
429 	}
430 
431 	return false;
432 }
433 
434 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
435 {
436 	return !(cgrp->self.flags & CSS_ONLINE);
437 }
438 
439 static inline void cgroup_get(struct cgroup *cgrp)
440 {
441 	css_get(&cgrp->self);
442 }
443 
444 static inline bool cgroup_tryget(struct cgroup *cgrp)
445 {
446 	return css_tryget(&cgrp->self);
447 }
448 
449 static inline void cgroup_put(struct cgroup *cgrp)
450 {
451 	css_put(&cgrp->self);
452 }
453 
454 static inline void cgroup_lock(void)
455 {
456 	mutex_lock(&cgroup_mutex);
457 }
458 
459 static inline void cgroup_unlock(void)
460 {
461 	mutex_unlock(&cgroup_mutex);
462 }
463 
464 /**
465  * task_css_set_check - obtain a task's css_set with extra access conditions
466  * @task: the task to obtain css_set for
467  * @__c: extra condition expression to be passed to rcu_dereference_check()
468  *
469  * A task's css_set is RCU protected, initialized and exited while holding
470  * task_lock(), and can only be modified while holding both cgroup_mutex
471  * and task_lock() while the task is alive.  This macro verifies that the
472  * caller is inside proper critical section and returns @task's css_set.
473  *
474  * The caller can also specify additional allowed conditions via @__c, such
475  * as locks used during the cgroup_subsys::attach() methods.
476  */
477 #ifdef CONFIG_PROVE_RCU
478 #define task_css_set_check(task, __c)					\
479 	rcu_dereference_check((task)->cgroups,				\
480 		rcu_read_lock_sched_held() ||				\
481 		lockdep_is_held(&cgroup_mutex) ||			\
482 		lockdep_is_held(&css_set_lock) ||			\
483 		((task)->flags & PF_EXITING) || (__c))
484 #else
485 #define task_css_set_check(task, __c)					\
486 	rcu_dereference((task)->cgroups)
487 #endif
488 
489 /**
490  * task_css_check - obtain css for (task, subsys) w/ extra access conds
491  * @task: the target task
492  * @subsys_id: the target subsystem ID
493  * @__c: extra condition expression to be passed to rcu_dereference_check()
494  *
495  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
496  * synchronization rules are the same as task_css_set_check().
497  */
498 #define task_css_check(task, subsys_id, __c)				\
499 	task_css_set_check((task), (__c))->subsys[(subsys_id)]
500 
501 /**
502  * task_css_set - obtain a task's css_set
503  * @task: the task to obtain css_set for
504  *
505  * See task_css_set_check().
506  */
507 static inline struct css_set *task_css_set(struct task_struct *task)
508 {
509 	return task_css_set_check(task, false);
510 }
511 
512 /**
513  * task_css - obtain css for (task, subsys)
514  * @task: the target task
515  * @subsys_id: the target subsystem ID
516  *
517  * See task_css_check().
518  */
519 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
520 						   int subsys_id)
521 {
522 	return task_css_check(task, subsys_id, false);
523 }
524 
525 /**
526  * task_get_css - find and get the css for (task, subsys)
527  * @task: the target task
528  * @subsys_id: the target subsystem ID
529  *
530  * Find the css for the (@task, @subsys_id) combination, increment a
531  * reference on and return it.  This function is guaranteed to return a
532  * valid css.  The returned css may already have been offlined.
533  */
534 static inline struct cgroup_subsys_state *
535 task_get_css(struct task_struct *task, int subsys_id)
536 {
537 	struct cgroup_subsys_state *css;
538 
539 	rcu_read_lock();
540 	while (true) {
541 		css = task_css(task, subsys_id);
542 		/*
543 		 * Can't use css_tryget_online() here.  A task which has
544 		 * PF_EXITING set may stay associated with an offline css.
545 		 * If such task calls this function, css_tryget_online()
546 		 * will keep failing.
547 		 */
548 		if (likely(css_tryget(css)))
549 			break;
550 		cpu_relax();
551 	}
552 	rcu_read_unlock();
553 	return css;
554 }
555 
556 /**
557  * task_css_is_root - test whether a task belongs to the root css
558  * @task: the target task
559  * @subsys_id: the target subsystem ID
560  *
561  * Test whether @task belongs to the root css on the specified subsystem.
562  * May be invoked in any context.
563  */
564 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
565 {
566 	return task_css_check(task, subsys_id, true) ==
567 		init_css_set.subsys[subsys_id];
568 }
569 
570 static inline struct cgroup *task_cgroup(struct task_struct *task,
571 					 int subsys_id)
572 {
573 	return task_css(task, subsys_id)->cgroup;
574 }
575 
576 static inline struct cgroup *task_dfl_cgroup(struct task_struct *task)
577 {
578 	return task_css_set(task)->dfl_cgrp;
579 }
580 
581 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
582 {
583 	struct cgroup_subsys_state *parent_css = cgrp->self.parent;
584 
585 	if (parent_css)
586 		return container_of(parent_css, struct cgroup, self);
587 	return NULL;
588 }
589 
590 /**
591  * cgroup_is_descendant - test ancestry
592  * @cgrp: the cgroup to be tested
593  * @ancestor: possible ancestor of @cgrp
594  *
595  * Test whether @cgrp is a descendant of @ancestor.  It also returns %true
596  * if @cgrp == @ancestor.  This function is safe to call as long as @cgrp
597  * and @ancestor are accessible.
598  */
599 static inline bool cgroup_is_descendant(struct cgroup *cgrp,
600 					struct cgroup *ancestor)
601 {
602 	if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
603 		return false;
604 	return cgrp->ancestors[ancestor->level] == ancestor;
605 }
606 
607 /**
608  * cgroup_ancestor - find ancestor of cgroup
609  * @cgrp: cgroup to find ancestor of
610  * @ancestor_level: level of ancestor to find starting from root
611  *
612  * Find ancestor of cgroup at specified level starting from root if it exists
613  * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at
614  * @ancestor_level.
615  *
616  * This function is safe to call as long as @cgrp is accessible.
617  */
618 static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp,
619 					     int ancestor_level)
620 {
621 	if (ancestor_level < 0 || ancestor_level > cgrp->level)
622 		return NULL;
623 	return cgrp->ancestors[ancestor_level];
624 }
625 
626 /**
627  * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
628  * @task: the task to be tested
629  * @ancestor: possible ancestor of @task's cgroup
630  *
631  * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
632  * It follows all the same rules as cgroup_is_descendant, and only applies
633  * to the default hierarchy.
634  */
635 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
636 					       struct cgroup *ancestor)
637 {
638 	struct css_set *cset = task_css_set(task);
639 
640 	return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
641 }
642 
643 /*
644  * Populated counters: writes happen under css_set_lock. The accessors below
645  * may read unlocked. What an unpopulated result means depends on context:
646  *
647  * - No lock held. Just a snapshot. May race with concurrent updates and is
648  *   useful only as a hint.
649  *
650  * - cgroup_mutex held. Migration into the cgroup is blocked, so an observed
651  *   !populated stays !populated until cgroup_mutex is dropped.
652  *
653  * - CSS_DYING set. The css can no longer be repopulated, so !populated is
654  *   sticky once observed.
655  */
656 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
657 {
658 	return READ_ONCE(cgrp->self.nr_populated_csets);
659 }
660 
661 static inline bool css_is_populated(struct cgroup_subsys_state *css)
662 {
663 	return READ_ONCE(css->nr_populated_csets) || READ_ONCE(css->nr_populated_children);
664 }
665 
666 static inline bool cgroup_is_populated(struct cgroup *cgrp)
667 {
668 	return css_is_populated(&cgrp->self);
669 }
670 
671 /* returns ino associated with a cgroup */
672 static inline ino_t cgroup_ino(struct cgroup *cgrp)
673 {
674 	return kernfs_ino(cgrp->kn);
675 }
676 
677 /* cft/css accessors for cftype->write() operation */
678 static inline struct cftype *of_cft(struct kernfs_open_file *of)
679 {
680 	return of->kn->priv;
681 }
682 
683 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
684 
685 /* cft/css accessors for cftype->seq_*() operations */
686 static inline struct cftype *seq_cft(struct seq_file *seq)
687 {
688 	return of_cft(seq->private);
689 }
690 
691 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
692 {
693 	return of_css(seq->private);
694 }
695 
696 /*
697  * Name / path handling functions.  All are thin wrappers around the kernfs
698  * counterparts and can be called under any context.
699  */
700 
701 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
702 {
703 	return kernfs_name(cgrp->kn, buf, buflen);
704 }
705 
706 static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
707 {
708 	return kernfs_path(cgrp->kn, buf, buflen);
709 }
710 
711 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
712 {
713 	pr_cont_kernfs_name(cgrp->kn);
714 }
715 
716 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
717 {
718 	pr_cont_kernfs_path(cgrp->kn);
719 }
720 
721 bool cgroup_psi_enabled(void);
722 
723 static inline void cgroup_init_kthreadd(void)
724 {
725 	/*
726 	 * kthreadd is inherited by all kthreads, keep it in the root so
727 	 * that the new kthreads are guaranteed to stay in the root until
728 	 * initialization is finished.
729 	 */
730 	current->no_cgroup_migration = 1;
731 }
732 
733 static inline void cgroup_kthread_ready(void)
734 {
735 	/*
736 	 * This kthread finished initialization.  The creator should have
737 	 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
738 	 */
739 	current->no_cgroup_migration = 0;
740 }
741 
742 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen);
743 struct cgroup *__cgroup_get_from_id(u64 id);
744 struct cgroup *cgroup_get_from_id(u64 id);
745 #else /* !CONFIG_CGROUPS */
746 
747 struct cgroup_subsys_state;
748 struct cgroup;
749 
750 static inline u64 cgroup_id(const struct cgroup *cgrp) { return 1; }
751 static inline void css_get(struct cgroup_subsys_state *css) {}
752 static inline void css_put(struct cgroup_subsys_state *css) {}
753 static inline void cgroup_lock(void) {}
754 static inline void cgroup_unlock(void) {}
755 static inline int cgroup_attach_task_all(struct task_struct *from,
756 					 struct task_struct *t) { return 0; }
757 static inline int cgroupstats_build(struct cgroupstats *stats,
758 				    struct dentry *dentry) { return -EINVAL; }
759 
760 static inline void cgroup_fork(struct task_struct *p) {}
761 static inline int cgroup_can_fork(struct task_struct *p,
762 				  struct kernel_clone_args *kargs) { return 0; }
763 static inline void cgroup_cancel_fork(struct task_struct *p,
764 				      struct kernel_clone_args *kargs) {}
765 static inline void cgroup_post_fork(struct task_struct *p,
766 				    struct kernel_clone_args *kargs) {}
767 static inline void cgroup_task_exit(struct task_struct *p) {}
768 static inline void cgroup_task_dead(struct task_struct *p) {}
769 static inline void cgroup_task_release(struct task_struct *p) {}
770 static inline void cgroup_task_free(struct task_struct *p) {}
771 
772 static inline int cgroup_init_early(void) { return 0; }
773 static inline int cgroup_init(void) { return 0; }
774 static inline void cgroup_init_kthreadd(void) {}
775 static inline void cgroup_kthread_ready(void) {}
776 
777 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp)
778 {
779 	return NULL;
780 }
781 
782 static inline bool cgroup_psi_enabled(void)
783 {
784 	return false;
785 }
786 
787 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
788 					       struct cgroup *ancestor)
789 {
790 	return true;
791 }
792 
793 static inline void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
794 {}
795 #endif /* !CONFIG_CGROUPS */
796 
797 #ifdef CONFIG_CGROUPS
798 /*
799  * cgroup scalable recursive statistics.
800  */
801 void __css_rstat_updated(struct cgroup_subsys_state *css, int cpu);
802 void css_rstat_updated(struct cgroup_subsys_state *css, int cpu);
803 void css_rstat_flush(struct cgroup_subsys_state *css);
804 
805 /*
806  * Basic resource stats.
807  */
808 #ifdef CONFIG_CGROUP_CPUACCT
809 void cpuacct_charge(struct task_struct *tsk, u64 cputime);
810 void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);
811 #else
812 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
813 static inline void cpuacct_account_field(struct task_struct *tsk, int index,
814 					 u64 val) {}
815 #endif
816 
817 void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec);
818 void __cgroup_account_cputime_field(struct cgroup *cgrp,
819 				    enum cpu_usage_stat index, u64 delta_exec);
820 
821 static inline void cgroup_account_cputime(struct task_struct *task,
822 					  u64 delta_exec)
823 {
824 	struct cgroup *cgrp;
825 
826 	cpuacct_charge(task, delta_exec);
827 
828 	cgrp = task_dfl_cgroup(task);
829 	if (cgroup_parent(cgrp))
830 		__cgroup_account_cputime(cgrp, delta_exec);
831 }
832 
833 static inline void cgroup_account_cputime_field(struct task_struct *task,
834 						enum cpu_usage_stat index,
835 						u64 delta_exec)
836 {
837 	struct cgroup *cgrp;
838 
839 	cpuacct_account_field(task, index, delta_exec);
840 
841 	cgrp = task_dfl_cgroup(task);
842 	if (cgroup_parent(cgrp))
843 		__cgroup_account_cputime_field(cgrp, index, delta_exec);
844 }
845 
846 #else	/* CONFIG_CGROUPS */
847 
848 static inline void cgroup_account_cputime(struct task_struct *task,
849 					  u64 delta_exec) {}
850 static inline void cgroup_account_cputime_field(struct task_struct *task,
851 						enum cpu_usage_stat index,
852 						u64 delta_exec) {}
853 
854 #endif	/* CONFIG_CGROUPS */
855 
856 /*
857  * sock->sk_cgrp_data handling.  For more info, see sock_cgroup_data
858  * definition in cgroup-defs.h.
859  */
860 #ifdef CONFIG_SOCK_CGROUP_DATA
861 
862 void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
863 void cgroup_sk_clone(struct sock_cgroup_data *skcd);
864 void cgroup_sk_free(struct sock_cgroup_data *skcd);
865 
866 static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
867 {
868 	return skcd->cgroup;
869 }
870 
871 #else	/* CONFIG_CGROUP_DATA */
872 
873 static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
874 static inline void cgroup_sk_clone(struct sock_cgroup_data *skcd) {}
875 static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
876 
877 #endif	/* CONFIG_CGROUP_DATA */
878 
879 #ifdef CONFIG_CGROUPS
880 
881 void cgroup_enter_frozen(void);
882 void cgroup_leave_frozen(bool always_leave);
883 void cgroup_update_frozen(struct cgroup *cgrp);
884 void cgroup_freeze(struct cgroup *cgrp, bool freeze);
885 void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src,
886 				 struct cgroup *dst);
887 
888 static inline bool cgroup_task_frozen(struct task_struct *task)
889 {
890 	return task->frozen;
891 }
892 
893 #else /* !CONFIG_CGROUPS */
894 
895 static inline void cgroup_enter_frozen(void) { }
896 static inline void cgroup_leave_frozen(bool always_leave) { }
897 static inline bool cgroup_task_frozen(struct task_struct *task)
898 {
899 	return false;
900 }
901 
902 #endif /* !CONFIG_CGROUPS */
903 
904 #ifdef CONFIG_CGROUP_BPF
905 static inline void cgroup_bpf_get(struct cgroup *cgrp)
906 {
907 	percpu_ref_get(&cgrp->bpf.refcnt);
908 }
909 
910 static inline void cgroup_bpf_put(struct cgroup *cgrp)
911 {
912 	percpu_ref_put(&cgrp->bpf.refcnt);
913 }
914 
915 #else /* CONFIG_CGROUP_BPF */
916 
917 static inline void cgroup_bpf_get(struct cgroup *cgrp) {}
918 static inline void cgroup_bpf_put(struct cgroup *cgrp) {}
919 
920 #endif /* CONFIG_CGROUP_BPF */
921 
922 struct cgroup *task_get_cgroup1(struct task_struct *tsk, int hierarchy_id);
923 
924 struct cgroup_of_peak *of_peak(struct kernfs_open_file *of);
925 
926 #endif /* _LINUX_CGROUP_H */
927