1 #ifndef _LINUX_CGROUP_H
2 #define _LINUX_CGROUP_H
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
11 #include <linux/sched.h>
12 #include <linux/cpumask.h>
13 #include <linux/nodemask.h>
14 #include <linux/rculist.h>
15 #include <linux/cgroupstats.h>
17 #include <linux/seq_file.h>
18 #include <linux/kernfs.h>
19 #include <linux/jump_label.h>
21 #include <linux/cgroup-defs.h>
26 * All weight knobs on the default hierarhcy should use the following min,
27 * default and max values. The default value is the logarithmic center of
28 * MIN and MAX and allows 100x to be expressed in both directions.
30 #define CGROUP_WEIGHT_MIN 1
31 #define CGROUP_WEIGHT_DFL 100
32 #define CGROUP_WEIGHT_MAX 10000
34 /* a css_task_iter should be treated as an opaque object */
35 struct css_task_iter {
36 struct cgroup_subsys *ss;
38 struct list_head *cset_pos;
39 struct list_head *cset_head;
41 struct list_head *task_pos;
42 struct list_head *tasks_head;
43 struct list_head *mg_tasks_head;
45 struct css_set *cur_cset;
46 struct task_struct *cur_task;
47 struct list_head iters_node; /* css_set->task_iters */
50 extern struct cgroup_root cgrp_dfl_root;
51 extern struct css_set init_css_set;
53 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
54 #include <linux/cgroup_subsys.h>
58 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \
59 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
60 #include <linux/cgroup_subsys.h>
64 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
65 * @ss: subsystem in question
67 #define cgroup_subsys_enabled(ss) \
68 static_branch_likely(&ss ## _enabled_key)
71 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
72 * @ss: subsystem in question
74 #define cgroup_subsys_on_dfl(ss) \
75 static_branch_likely(&ss ## _on_dfl_key)
77 bool css_has_online_children(struct cgroup_subsys_state *css);
78 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
79 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
80 struct cgroup_subsys *ss);
81 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
82 struct cgroup_subsys *ss);
84 struct cgroup *cgroup_get_from_path(const char *path);
86 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
87 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
89 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
90 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
91 int cgroup_rm_cftypes(struct cftype *cfts);
92 void cgroup_file_notify(struct cgroup_file *cfile);
94 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
95 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
96 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
97 struct pid *pid, struct task_struct *tsk);
99 void cgroup_fork(struct task_struct *p);
100 extern int cgroup_can_fork(struct task_struct *p);
101 extern void cgroup_cancel_fork(struct task_struct *p);
102 extern void cgroup_post_fork(struct task_struct *p);
103 void cgroup_exit(struct task_struct *p);
104 void cgroup_free(struct task_struct *p);
106 int cgroup_init_early(void);
107 int cgroup_init(void);
110 * Iteration helpers and macros.
113 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
114 struct cgroup_subsys_state *parent);
115 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
116 struct cgroup_subsys_state *css);
117 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
118 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
119 struct cgroup_subsys_state *css);
121 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
122 struct cgroup_subsys_state **dst_cssp);
123 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
124 struct cgroup_subsys_state **dst_cssp);
126 void css_task_iter_start(struct cgroup_subsys_state *css,
127 struct css_task_iter *it);
128 struct task_struct *css_task_iter_next(struct css_task_iter *it);
129 void css_task_iter_end(struct css_task_iter *it);
132 * css_for_each_child - iterate through children of a css
133 * @pos: the css * to use as the loop cursor
134 * @parent: css whose children to walk
136 * Walk @parent's children. Must be called under rcu_read_lock().
138 * If a subsystem synchronizes ->css_online() and the start of iteration, a
139 * css which finished ->css_online() is guaranteed to be visible in the
140 * future iterations and will stay visible until the last reference is put.
141 * A css which hasn't finished ->css_online() or already finished
142 * ->css_offline() may show up during traversal. It's each subsystem's
143 * responsibility to synchronize against on/offlining.
145 * It is allowed to temporarily drop RCU read lock during iteration. The
146 * caller is responsible for ensuring that @pos remains accessible until
147 * the start of the next iteration by, for example, bumping the css refcnt.
149 #define css_for_each_child(pos, parent) \
150 for ((pos) = css_next_child(NULL, (parent)); (pos); \
151 (pos) = css_next_child((pos), (parent)))
154 * css_for_each_descendant_pre - pre-order walk of a css's descendants
155 * @pos: the css * to use as the loop cursor
156 * @root: css whose descendants to walk
158 * Walk @root's descendants. @root is included in the iteration and the
159 * first node to be visited. Must be called under rcu_read_lock().
161 * If a subsystem synchronizes ->css_online() and the start of iteration, a
162 * css which finished ->css_online() is guaranteed to be visible in the
163 * future iterations and will stay visible until the last reference is put.
164 * A css which hasn't finished ->css_online() or already finished
165 * ->css_offline() may show up during traversal. It's each subsystem's
166 * responsibility to synchronize against on/offlining.
168 * For example, the following guarantees that a descendant can't escape
169 * state updates of its ancestors.
173 * Lock @css's parent and @css;
174 * Inherit state from the parent;
178 * my_update_state(@css)
180 * css_for_each_descendant_pre(@pos, @css) {
183 * Update @css's state;
185 * Verify @pos is alive and inherit state from its parent;
190 * As long as the inheriting step, including checking the parent state, is
191 * enclosed inside @pos locking, double-locking the parent isn't necessary
192 * while inheriting. The state update to the parent is guaranteed to be
193 * visible by walking order and, as long as inheriting operations to the
194 * same @pos are atomic to each other, multiple updates racing each other
195 * still result in the correct state. It's guaranateed that at least one
196 * inheritance happens for any css after the latest update to its parent.
198 * If checking parent's state requires locking the parent, each inheriting
199 * iteration should lock and unlock both @pos->parent and @pos.
201 * Alternatively, a subsystem may choose to use a single global lock to
202 * synchronize ->css_online() and ->css_offline() against tree-walking
205 * It is allowed to temporarily drop RCU read lock during iteration. The
206 * caller is responsible for ensuring that @pos remains accessible until
207 * the start of the next iteration by, for example, bumping the css refcnt.
209 #define css_for_each_descendant_pre(pos, css) \
210 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
211 (pos) = css_next_descendant_pre((pos), (css)))
214 * css_for_each_descendant_post - post-order walk of a css's descendants
215 * @pos: the css * to use as the loop cursor
216 * @css: css whose descendants to walk
218 * Similar to css_for_each_descendant_pre() but performs post-order
219 * traversal instead. @root is included in the iteration and the last
220 * node to be visited.
222 * If a subsystem synchronizes ->css_online() and the start of iteration, a
223 * css which finished ->css_online() is guaranteed to be visible in the
224 * future iterations and will stay visible until the last reference is put.
225 * A css which hasn't finished ->css_online() or already finished
226 * ->css_offline() may show up during traversal. It's each subsystem's
227 * responsibility to synchronize against on/offlining.
229 * Note that the walk visibility guarantee example described in pre-order
230 * walk doesn't apply the same to post-order walks.
232 #define css_for_each_descendant_post(pos, css) \
233 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
234 (pos) = css_next_descendant_post((pos), (css)))
237 * cgroup_taskset_for_each - iterate cgroup_taskset
238 * @task: the loop cursor
239 * @dst_css: the destination css
240 * @tset: taskset to iterate
242 * @tset may contain multiple tasks and they may belong to multiple
245 * On the v2 hierarchy, there may be tasks from multiple processes and they
246 * may not share the source or destination csses.
248 * On traditional hierarchies, when there are multiple tasks in @tset, if a
249 * task of a process is in @tset, all tasks of the process are in @tset.
250 * Also, all are guaranteed to share the same source and destination csses.
252 * Iteration is not in any specific order.
254 #define cgroup_taskset_for_each(task, dst_css, tset) \
255 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \
257 (task) = cgroup_taskset_next((tset), &(dst_css)))
260 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
261 * @leader: the loop cursor
262 * @dst_css: the destination css
263 * @tset: takset to iterate
265 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
266 * may not contain any.
268 #define cgroup_taskset_for_each_leader(leader, dst_css, tset) \
269 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \
271 (leader) = cgroup_taskset_next((tset), &(dst_css))) \
272 if ((leader) != (leader)->group_leader) \
281 * css_get - obtain a reference on the specified css
284 * The caller must already have a reference.
286 static inline void css_get(struct cgroup_subsys_state *css)
288 if (!(css->flags & CSS_NO_REF))
289 percpu_ref_get(&css->refcnt);
293 * css_get_many - obtain references on the specified css
295 * @n: number of references to get
297 * The caller must already have a reference.
299 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
301 if (!(css->flags & CSS_NO_REF))
302 percpu_ref_get_many(&css->refcnt, n);
306 * css_tryget - try to obtain a reference on the specified css
309 * Obtain a reference on @css unless it already has reached zero and is
310 * being released. This function doesn't care whether @css is on or
311 * offline. The caller naturally needs to ensure that @css is accessible
312 * but doesn't have to be holding a reference on it - IOW, RCU protected
313 * access is good enough for this function. Returns %true if a reference
314 * count was successfully obtained; %false otherwise.
316 static inline bool css_tryget(struct cgroup_subsys_state *css)
318 if (!(css->flags & CSS_NO_REF))
319 return percpu_ref_tryget(&css->refcnt);
324 * css_tryget_online - try to obtain a reference on the specified css if online
327 * Obtain a reference on @css if it's online. The caller naturally needs
328 * to ensure that @css is accessible but doesn't have to be holding a
329 * reference on it - IOW, RCU protected access is good enough for this
330 * function. Returns %true if a reference count was successfully obtained;
333 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
335 if (!(css->flags & CSS_NO_REF))
336 return percpu_ref_tryget_live(&css->refcnt);
341 * css_put - put a css reference
344 * Put a reference obtained via css_get() and css_tryget_online().
346 static inline void css_put(struct cgroup_subsys_state *css)
348 if (!(css->flags & CSS_NO_REF))
349 percpu_ref_put(&css->refcnt);
353 * css_put_many - put css references
355 * @n: number of references to put
357 * Put references obtained via css_get() and css_tryget_online().
359 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
361 if (!(css->flags & CSS_NO_REF))
362 percpu_ref_put_many(&css->refcnt, n);
365 static inline void cgroup_put(struct cgroup *cgrp)
367 css_put(&cgrp->self);
371 * task_css_set_check - obtain a task's css_set with extra access conditions
372 * @task: the task to obtain css_set for
373 * @__c: extra condition expression to be passed to rcu_dereference_check()
375 * A task's css_set is RCU protected, initialized and exited while holding
376 * task_lock(), and can only be modified while holding both cgroup_mutex
377 * and task_lock() while the task is alive. This macro verifies that the
378 * caller is inside proper critical section and returns @task's css_set.
380 * The caller can also specify additional allowed conditions via @__c, such
381 * as locks used during the cgroup_subsys::attach() methods.
383 #ifdef CONFIG_PROVE_RCU
384 extern struct mutex cgroup_mutex;
385 extern spinlock_t css_set_lock;
386 #define task_css_set_check(task, __c) \
387 rcu_dereference_check((task)->cgroups, \
388 lockdep_is_held(&cgroup_mutex) || \
389 lockdep_is_held(&css_set_lock) || \
390 ((task)->flags & PF_EXITING) || (__c))
392 #define task_css_set_check(task, __c) \
393 rcu_dereference((task)->cgroups)
397 * task_css_check - obtain css for (task, subsys) w/ extra access conds
398 * @task: the target task
399 * @subsys_id: the target subsystem ID
400 * @__c: extra condition expression to be passed to rcu_dereference_check()
402 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
403 * synchronization rules are the same as task_css_set_check().
405 #define task_css_check(task, subsys_id, __c) \
406 task_css_set_check((task), (__c))->subsys[(subsys_id)]
409 * task_css_set - obtain a task's css_set
410 * @task: the task to obtain css_set for
412 * See task_css_set_check().
414 static inline struct css_set *task_css_set(struct task_struct *task)
416 return task_css_set_check(task, false);
420 * task_css - obtain css for (task, subsys)
421 * @task: the target task
422 * @subsys_id: the target subsystem ID
424 * See task_css_check().
426 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
429 return task_css_check(task, subsys_id, false);
433 * task_get_css - find and get the css for (task, subsys)
434 * @task: the target task
435 * @subsys_id: the target subsystem ID
437 * Find the css for the (@task, @subsys_id) combination, increment a
438 * reference on and return it. This function is guaranteed to return a
441 static inline struct cgroup_subsys_state *
442 task_get_css(struct task_struct *task, int subsys_id)
444 struct cgroup_subsys_state *css;
448 css = task_css(task, subsys_id);
449 if (likely(css_tryget_online(css)))
458 * task_css_is_root - test whether a task belongs to the root css
459 * @task: the target task
460 * @subsys_id: the target subsystem ID
462 * Test whether @task belongs to the root css on the specified subsystem.
463 * May be invoked in any context.
465 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
467 return task_css_check(task, subsys_id, true) ==
468 init_css_set.subsys[subsys_id];
471 static inline struct cgroup *task_cgroup(struct task_struct *task,
474 return task_css(task, subsys_id)->cgroup;
478 * cgroup_is_descendant - test ancestry
479 * @cgrp: the cgroup to be tested
480 * @ancestor: possible ancestor of @cgrp
482 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
483 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
484 * and @ancestor are accessible.
486 static inline bool cgroup_is_descendant(struct cgroup *cgrp,
487 struct cgroup *ancestor)
489 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
491 return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
494 /* no synchronization, the result can only be used as a hint */
495 static inline bool cgroup_is_populated(struct cgroup *cgrp)
497 return cgrp->populated_cnt;
500 /* returns ino associated with a cgroup */
501 static inline ino_t cgroup_ino(struct cgroup *cgrp)
503 return cgrp->kn->ino;
506 /* cft/css accessors for cftype->write() operation */
507 static inline struct cftype *of_cft(struct kernfs_open_file *of)
512 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
514 /* cft/css accessors for cftype->seq_*() operations */
515 static inline struct cftype *seq_cft(struct seq_file *seq)
517 return of_cft(seq->private);
520 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
522 return of_css(seq->private);
526 * Name / path handling functions. All are thin wrappers around the kernfs
527 * counterparts and can be called under any context.
530 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
532 return kernfs_name(cgrp->kn, buf, buflen);
535 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
538 return kernfs_path(cgrp->kn, buf, buflen);
541 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
543 pr_cont_kernfs_name(cgrp->kn);
546 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
548 pr_cont_kernfs_path(cgrp->kn);
551 #else /* !CONFIG_CGROUPS */
553 struct cgroup_subsys_state;
555 static inline void css_put(struct cgroup_subsys_state *css) {}
556 static inline int cgroup_attach_task_all(struct task_struct *from,
557 struct task_struct *t) { return 0; }
558 static inline int cgroupstats_build(struct cgroupstats *stats,
559 struct dentry *dentry) { return -EINVAL; }
561 static inline void cgroup_fork(struct task_struct *p) {}
562 static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
563 static inline void cgroup_cancel_fork(struct task_struct *p) {}
564 static inline void cgroup_post_fork(struct task_struct *p) {}
565 static inline void cgroup_exit(struct task_struct *p) {}
566 static inline void cgroup_free(struct task_struct *p) {}
568 static inline int cgroup_init_early(void) { return 0; }
569 static inline int cgroup_init(void) { return 0; }
571 #endif /* !CONFIG_CGROUPS */
574 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
575 * definition in cgroup-defs.h.
577 #ifdef CONFIG_SOCK_CGROUP_DATA
579 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
580 extern spinlock_t cgroup_sk_update_lock;
583 void cgroup_sk_alloc_disable(void);
584 void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
585 void cgroup_sk_free(struct sock_cgroup_data *skcd);
587 static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
589 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
593 * @skcd->val is 64bit but the following is safe on 32bit too as we
594 * just need the lower ulong to be written and read atomically.
596 v = READ_ONCE(skcd->val);
599 return &cgrp_dfl_root.cgrp;
601 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
603 return (struct cgroup *)(unsigned long)skcd->val;
607 #else /* CONFIG_CGROUP_DATA */
609 static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
610 static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
612 #endif /* CONFIG_CGROUP_DATA */
614 #endif /* _LINUX_CGROUP_H */