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 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
85 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
86 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
88 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
89 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
90 int cgroup_rm_cftypes(struct cftype *cfts);
91 void cgroup_file_notify(struct cgroup_file *cfile);
93 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
94 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
95 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
96 struct pid *pid, struct task_struct *tsk);
98 void cgroup_fork(struct task_struct *p);
99 extern int cgroup_can_fork(struct task_struct *p,
100 void *ss_priv[CGROUP_CANFORK_COUNT]);
101 extern void cgroup_cancel_fork(struct task_struct *p,
102 void *ss_priv[CGROUP_CANFORK_COUNT]);
103 extern void cgroup_post_fork(struct task_struct *p,
104 void *old_ss_priv[CGROUP_CANFORK_COUNT]);
105 void cgroup_exit(struct task_struct *p);
106 void cgroup_free(struct task_struct *p);
108 int cgroup_init_early(void);
109 int cgroup_init(void);
112 * Iteration helpers and macros.
115 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
116 struct cgroup_subsys_state *parent);
117 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
118 struct cgroup_subsys_state *css);
119 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
120 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
121 struct cgroup_subsys_state *css);
123 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
124 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
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 * @tset: taskset to iterate
241 * @tset may contain multiple tasks and they may belong to multiple
242 * processes. When there are multiple tasks in @tset, if a task of a
243 * process is in @tset, all tasks of the process are in @tset. Also, all
244 * are guaranteed to share the same source and destination csses.
246 * Iteration is not in any specific order.
248 #define cgroup_taskset_for_each(task, tset) \
249 for ((task) = cgroup_taskset_first((tset)); (task); \
250 (task) = cgroup_taskset_next((tset)))
253 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
254 * @leader: the loop cursor
255 * @tset: takset to iterate
257 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
258 * may not contain any.
260 #define cgroup_taskset_for_each_leader(leader, tset) \
261 for ((leader) = cgroup_taskset_first((tset)); (leader); \
262 (leader) = cgroup_taskset_next((tset))) \
263 if ((leader) != (leader)->group_leader) \
272 * css_get - obtain a reference on the specified css
275 * The caller must already have a reference.
277 static inline void css_get(struct cgroup_subsys_state *css)
279 if (!(css->flags & CSS_NO_REF))
280 percpu_ref_get(&css->refcnt);
284 * css_get_many - obtain references on the specified css
286 * @n: number of references to get
288 * The caller must already have a reference.
290 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
292 if (!(css->flags & CSS_NO_REF))
293 percpu_ref_get_many(&css->refcnt, n);
297 * css_tryget - try to obtain a reference on the specified css
300 * Obtain a reference on @css unless it already has reached zero and is
301 * being released. This function doesn't care whether @css is on or
302 * offline. The caller naturally needs to ensure that @css is accessible
303 * but doesn't have to be holding a reference on it - IOW, RCU protected
304 * access is good enough for this function. Returns %true if a reference
305 * count was successfully obtained; %false otherwise.
307 static inline bool css_tryget(struct cgroup_subsys_state *css)
309 if (!(css->flags & CSS_NO_REF))
310 return percpu_ref_tryget(&css->refcnt);
315 * css_tryget_online - try to obtain a reference on the specified css if online
318 * Obtain a reference on @css if it's online. The caller naturally needs
319 * to ensure that @css is accessible but doesn't have to be holding a
320 * reference on it - IOW, RCU protected access is good enough for this
321 * function. Returns %true if a reference count was successfully obtained;
324 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
326 if (!(css->flags & CSS_NO_REF))
327 return percpu_ref_tryget_live(&css->refcnt);
332 * css_put - put a css reference
335 * Put a reference obtained via css_get() and css_tryget_online().
337 static inline void css_put(struct cgroup_subsys_state *css)
339 if (!(css->flags & CSS_NO_REF))
340 percpu_ref_put(&css->refcnt);
344 * css_put_many - put css references
346 * @n: number of references to put
348 * Put references obtained via css_get() and css_tryget_online().
350 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
352 if (!(css->flags & CSS_NO_REF))
353 percpu_ref_put_many(&css->refcnt, n);
357 * task_css_set_check - obtain a task's css_set with extra access conditions
358 * @task: the task to obtain css_set for
359 * @__c: extra condition expression to be passed to rcu_dereference_check()
361 * A task's css_set is RCU protected, initialized and exited while holding
362 * task_lock(), and can only be modified while holding both cgroup_mutex
363 * and task_lock() while the task is alive. This macro verifies that the
364 * caller is inside proper critical section and returns @task's css_set.
366 * The caller can also specify additional allowed conditions via @__c, such
367 * as locks used during the cgroup_subsys::attach() methods.
369 #ifdef CONFIG_PROVE_RCU
370 extern struct mutex cgroup_mutex;
371 extern spinlock_t css_set_lock;
372 #define task_css_set_check(task, __c) \
373 rcu_dereference_check((task)->cgroups, \
374 lockdep_is_held(&cgroup_mutex) || \
375 lockdep_is_held(&css_set_lock) || \
376 ((task)->flags & PF_EXITING) || (__c))
378 #define task_css_set_check(task, __c) \
379 rcu_dereference((task)->cgroups)
383 * task_css_check - obtain css for (task, subsys) w/ extra access conds
384 * @task: the target task
385 * @subsys_id: the target subsystem ID
386 * @__c: extra condition expression to be passed to rcu_dereference_check()
388 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
389 * synchronization rules are the same as task_css_set_check().
391 #define task_css_check(task, subsys_id, __c) \
392 task_css_set_check((task), (__c))->subsys[(subsys_id)]
395 * task_css_set - obtain a task's css_set
396 * @task: the task to obtain css_set for
398 * See task_css_set_check().
400 static inline struct css_set *task_css_set(struct task_struct *task)
402 return task_css_set_check(task, false);
406 * task_css - obtain css for (task, subsys)
407 * @task: the target task
408 * @subsys_id: the target subsystem ID
410 * See task_css_check().
412 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
415 return task_css_check(task, subsys_id, false);
419 * task_get_css - find and get the css for (task, subsys)
420 * @task: the target task
421 * @subsys_id: the target subsystem ID
423 * Find the css for the (@task, @subsys_id) combination, increment a
424 * reference on and return it. This function is guaranteed to return a
427 static inline struct cgroup_subsys_state *
428 task_get_css(struct task_struct *task, int subsys_id)
430 struct cgroup_subsys_state *css;
434 css = task_css(task, subsys_id);
435 if (likely(css_tryget_online(css)))
444 * task_css_is_root - test whether a task belongs to the root css
445 * @task: the target task
446 * @subsys_id: the target subsystem ID
448 * Test whether @task belongs to the root css on the specified subsystem.
449 * May be invoked in any context.
451 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
453 return task_css_check(task, subsys_id, true) ==
454 init_css_set.subsys[subsys_id];
457 static inline struct cgroup *task_cgroup(struct task_struct *task,
460 return task_css(task, subsys_id)->cgroup;
463 /* no synchronization, the result can only be used as a hint */
464 static inline bool cgroup_is_populated(struct cgroup *cgrp)
466 return cgrp->populated_cnt;
469 /* returns ino associated with a cgroup */
470 static inline ino_t cgroup_ino(struct cgroup *cgrp)
472 return cgrp->kn->ino;
475 /* cft/css accessors for cftype->write() operation */
476 static inline struct cftype *of_cft(struct kernfs_open_file *of)
481 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
483 /* cft/css accessors for cftype->seq_*() operations */
484 static inline struct cftype *seq_cft(struct seq_file *seq)
486 return of_cft(seq->private);
489 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
491 return of_css(seq->private);
495 * Name / path handling functions. All are thin wrappers around the kernfs
496 * counterparts and can be called under any context.
499 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
501 return kernfs_name(cgrp->kn, buf, buflen);
504 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
507 return kernfs_path(cgrp->kn, buf, buflen);
510 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
512 pr_cont_kernfs_name(cgrp->kn);
515 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
517 pr_cont_kernfs_path(cgrp->kn);
520 #else /* !CONFIG_CGROUPS */
522 struct cgroup_subsys_state;
524 static inline void css_put(struct cgroup_subsys_state *css) {}
525 static inline int cgroup_attach_task_all(struct task_struct *from,
526 struct task_struct *t) { return 0; }
527 static inline int cgroupstats_build(struct cgroupstats *stats,
528 struct dentry *dentry) { return -EINVAL; }
530 static inline void cgroup_fork(struct task_struct *p) {}
531 static inline int cgroup_can_fork(struct task_struct *p,
532 void *ss_priv[CGROUP_CANFORK_COUNT])
534 static inline void cgroup_cancel_fork(struct task_struct *p,
535 void *ss_priv[CGROUP_CANFORK_COUNT]) {}
536 static inline void cgroup_post_fork(struct task_struct *p,
537 void *ss_priv[CGROUP_CANFORK_COUNT]) {}
538 static inline void cgroup_exit(struct task_struct *p) {}
539 static inline void cgroup_free(struct task_struct *p) {}
541 static inline int cgroup_init_early(void) { return 0; }
542 static inline int cgroup_init(void) { return 0; }
544 #endif /* !CONFIG_CGROUPS */
546 #endif /* _LINUX_CGROUP_H */