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>
20 #include <linux/types.h>
21 #include <linux/ns_common.h>
22 #include <linux/nsproxy.h>
23 #include <linux/user_namespace.h>
24 #include <linux/refcount.h>
26 #include <linux/cgroup-defs.h>
31 * All weight knobs on the default hierarhcy should use the following min,
32 * default and max values. The default value is the logarithmic center of
33 * MIN and MAX and allows 100x to be expressed in both directions.
35 #define CGROUP_WEIGHT_MIN 1
36 #define CGROUP_WEIGHT_DFL 100
37 #define CGROUP_WEIGHT_MAX 10000
39 /* a css_task_iter should be treated as an opaque object */
40 struct css_task_iter {
41 struct cgroup_subsys *ss;
43 struct list_head *cset_pos;
44 struct list_head *cset_head;
46 struct list_head *task_pos;
47 struct list_head *tasks_head;
48 struct list_head *mg_tasks_head;
50 struct css_set *cur_cset;
51 struct task_struct *cur_task;
52 struct list_head iters_node; /* css_set->task_iters */
55 extern struct cgroup_root cgrp_dfl_root;
56 extern struct css_set init_css_set;
58 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
59 #include <linux/cgroup_subsys.h>
63 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \
64 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key;
65 #include <linux/cgroup_subsys.h>
69 * cgroup_subsys_enabled - fast test on whether a subsys is enabled
70 * @ss: subsystem in question
72 #define cgroup_subsys_enabled(ss) \
73 static_branch_likely(&ss ## _enabled_key)
76 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy
77 * @ss: subsystem in question
79 #define cgroup_subsys_on_dfl(ss) \
80 static_branch_likely(&ss ## _on_dfl_key)
82 bool css_has_online_children(struct cgroup_subsys_state *css);
83 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
84 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
85 struct cgroup_subsys *ss);
86 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
87 struct cgroup_subsys *ss);
89 struct cgroup *cgroup_get_from_path(const char *path);
90 struct cgroup *cgroup_get_from_fd(int fd);
92 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
93 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
95 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
96 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
97 int cgroup_rm_cftypes(struct cftype *cfts);
98 void cgroup_file_notify(struct cgroup_file *cfile);
100 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
101 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
102 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
103 struct pid *pid, struct task_struct *tsk);
105 void cgroup_fork(struct task_struct *p);
106 extern int cgroup_can_fork(struct task_struct *p);
107 extern void cgroup_cancel_fork(struct task_struct *p);
108 extern void cgroup_post_fork(struct task_struct *p);
109 void cgroup_exit(struct task_struct *p);
110 void cgroup_free(struct task_struct *p);
112 int cgroup_init_early(void);
113 int cgroup_init(void);
116 * Iteration helpers and macros.
119 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
120 struct cgroup_subsys_state *parent);
121 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
122 struct cgroup_subsys_state *css);
123 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
124 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
125 struct cgroup_subsys_state *css);
127 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
128 struct cgroup_subsys_state **dst_cssp);
129 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
130 struct cgroup_subsys_state **dst_cssp);
132 void css_task_iter_start(struct cgroup_subsys_state *css,
133 struct css_task_iter *it);
134 struct task_struct *css_task_iter_next(struct css_task_iter *it);
135 void css_task_iter_end(struct css_task_iter *it);
138 * css_for_each_child - iterate through children of a css
139 * @pos: the css * to use as the loop cursor
140 * @parent: css whose children to walk
142 * Walk @parent's children. Must be called under rcu_read_lock().
144 * If a subsystem synchronizes ->css_online() and the start of iteration, a
145 * css which finished ->css_online() is guaranteed to be visible in the
146 * future iterations and will stay visible until the last reference is put.
147 * A css which hasn't finished ->css_online() or already finished
148 * ->css_offline() may show up during traversal. It's each subsystem's
149 * responsibility to synchronize against on/offlining.
151 * It is allowed to temporarily drop RCU read lock during iteration. The
152 * caller is responsible for ensuring that @pos remains accessible until
153 * the start of the next iteration by, for example, bumping the css refcnt.
155 #define css_for_each_child(pos, parent) \
156 for ((pos) = css_next_child(NULL, (parent)); (pos); \
157 (pos) = css_next_child((pos), (parent)))
160 * css_for_each_descendant_pre - pre-order walk of a css's descendants
161 * @pos: the css * to use as the loop cursor
162 * @root: css whose descendants to walk
164 * Walk @root's descendants. @root is included in the iteration and the
165 * first node to be visited. Must be called under rcu_read_lock().
167 * If a subsystem synchronizes ->css_online() and the start of iteration, a
168 * css which finished ->css_online() is guaranteed to be visible in the
169 * future iterations and will stay visible until the last reference is put.
170 * A css which hasn't finished ->css_online() or already finished
171 * ->css_offline() may show up during traversal. It's each subsystem's
172 * responsibility to synchronize against on/offlining.
174 * For example, the following guarantees that a descendant can't escape
175 * state updates of its ancestors.
179 * Lock @css's parent and @css;
180 * Inherit state from the parent;
184 * my_update_state(@css)
186 * css_for_each_descendant_pre(@pos, @css) {
189 * Update @css's state;
191 * Verify @pos is alive and inherit state from its parent;
196 * As long as the inheriting step, including checking the parent state, is
197 * enclosed inside @pos locking, double-locking the parent isn't necessary
198 * while inheriting. The state update to the parent is guaranteed to be
199 * visible by walking order and, as long as inheriting operations to the
200 * same @pos are atomic to each other, multiple updates racing each other
201 * still result in the correct state. It's guaranateed that at least one
202 * inheritance happens for any css after the latest update to its parent.
204 * If checking parent's state requires locking the parent, each inheriting
205 * iteration should lock and unlock both @pos->parent and @pos.
207 * Alternatively, a subsystem may choose to use a single global lock to
208 * synchronize ->css_online() and ->css_offline() against tree-walking
211 * It is allowed to temporarily drop RCU read lock during iteration. The
212 * caller is responsible for ensuring that @pos remains accessible until
213 * the start of the next iteration by, for example, bumping the css refcnt.
215 #define css_for_each_descendant_pre(pos, css) \
216 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
217 (pos) = css_next_descendant_pre((pos), (css)))
220 * css_for_each_descendant_post - post-order walk of a css's descendants
221 * @pos: the css * to use as the loop cursor
222 * @css: css whose descendants to walk
224 * Similar to css_for_each_descendant_pre() but performs post-order
225 * traversal instead. @root is included in the iteration and the last
226 * node to be visited.
228 * If a subsystem synchronizes ->css_online() and the start of iteration, a
229 * css which finished ->css_online() is guaranteed to be visible in the
230 * future iterations and will stay visible until the last reference is put.
231 * A css which hasn't finished ->css_online() or already finished
232 * ->css_offline() may show up during traversal. It's each subsystem's
233 * responsibility to synchronize against on/offlining.
235 * Note that the walk visibility guarantee example described in pre-order
236 * walk doesn't apply the same to post-order walks.
238 #define css_for_each_descendant_post(pos, css) \
239 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
240 (pos) = css_next_descendant_post((pos), (css)))
243 * cgroup_taskset_for_each - iterate cgroup_taskset
244 * @task: the loop cursor
245 * @dst_css: the destination css
246 * @tset: taskset to iterate
248 * @tset may contain multiple tasks and they may belong to multiple
251 * On the v2 hierarchy, there may be tasks from multiple processes and they
252 * may not share the source or destination csses.
254 * On traditional hierarchies, when there are multiple tasks in @tset, if a
255 * task of a process is in @tset, all tasks of the process are in @tset.
256 * Also, all are guaranteed to share the same source and destination csses.
258 * Iteration is not in any specific order.
260 #define cgroup_taskset_for_each(task, dst_css, tset) \
261 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \
263 (task) = cgroup_taskset_next((tset), &(dst_css)))
266 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset
267 * @leader: the loop cursor
268 * @dst_css: the destination css
269 * @tset: taskset to iterate
271 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset
272 * may not contain any.
274 #define cgroup_taskset_for_each_leader(leader, dst_css, tset) \
275 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \
277 (leader) = cgroup_taskset_next((tset), &(dst_css))) \
278 if ((leader) != (leader)->group_leader) \
287 * css_get - obtain a reference on the specified css
290 * The caller must already have a reference.
292 static inline void css_get(struct cgroup_subsys_state *css)
294 if (!(css->flags & CSS_NO_REF))
295 percpu_ref_get(&css->refcnt);
299 * css_get_many - obtain references on the specified css
301 * @n: number of references to get
303 * The caller must already have a reference.
305 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
307 if (!(css->flags & CSS_NO_REF))
308 percpu_ref_get_many(&css->refcnt, n);
312 * css_tryget - try to obtain a reference on the specified css
315 * Obtain a reference on @css unless it already has reached zero and is
316 * being released. This function doesn't care whether @css is on or
317 * offline. The caller naturally needs to ensure that @css is accessible
318 * but doesn't have to be holding a reference on it - IOW, RCU protected
319 * access is good enough for this function. Returns %true if a reference
320 * count was successfully obtained; %false otherwise.
322 static inline bool css_tryget(struct cgroup_subsys_state *css)
324 if (!(css->flags & CSS_NO_REF))
325 return percpu_ref_tryget(&css->refcnt);
330 * css_tryget_online - try to obtain a reference on the specified css if online
333 * Obtain a reference on @css if it's online. The caller naturally needs
334 * to ensure that @css is accessible but doesn't have to be holding a
335 * reference on it - IOW, RCU protected access is good enough for this
336 * function. Returns %true if a reference count was successfully obtained;
339 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
341 if (!(css->flags & CSS_NO_REF))
342 return percpu_ref_tryget_live(&css->refcnt);
347 * css_is_dying - test whether the specified css is dying
350 * Test whether @css is in the process of offlining or already offline. In
351 * most cases, ->css_online() and ->css_offline() callbacks should be
352 * enough; however, the actual offline operations are RCU delayed and this
353 * test returns %true also when @css is scheduled to be offlined.
355 * This is useful, for example, when the use case requires synchronous
356 * behavior with respect to cgroup removal. cgroup removal schedules css
357 * offlining but the css can seem alive while the operation is being
358 * delayed. If the delay affects user visible semantics, this test can be
359 * used to resolve the situation.
361 static inline bool css_is_dying(struct cgroup_subsys_state *css)
363 return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt);
367 * css_put - put a css reference
370 * Put a reference obtained via css_get() and css_tryget_online().
372 static inline void css_put(struct cgroup_subsys_state *css)
374 if (!(css->flags & CSS_NO_REF))
375 percpu_ref_put(&css->refcnt);
379 * css_put_many - put css references
381 * @n: number of references to put
383 * Put references obtained via css_get() and css_tryget_online().
385 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
387 if (!(css->flags & CSS_NO_REF))
388 percpu_ref_put_many(&css->refcnt, n);
391 static inline void cgroup_put(struct cgroup *cgrp)
393 css_put(&cgrp->self);
397 * task_css_set_check - obtain a task's css_set with extra access conditions
398 * @task: the task to obtain css_set for
399 * @__c: extra condition expression to be passed to rcu_dereference_check()
401 * A task's css_set is RCU protected, initialized and exited while holding
402 * task_lock(), and can only be modified while holding both cgroup_mutex
403 * and task_lock() while the task is alive. This macro verifies that the
404 * caller is inside proper critical section and returns @task's css_set.
406 * The caller can also specify additional allowed conditions via @__c, such
407 * as locks used during the cgroup_subsys::attach() methods.
409 #ifdef CONFIG_PROVE_RCU
410 extern struct mutex cgroup_mutex;
411 extern spinlock_t css_set_lock;
412 #define task_css_set_check(task, __c) \
413 rcu_dereference_check((task)->cgroups, \
414 lockdep_is_held(&cgroup_mutex) || \
415 lockdep_is_held(&css_set_lock) || \
416 ((task)->flags & PF_EXITING) || (__c))
418 #define task_css_set_check(task, __c) \
419 rcu_dereference((task)->cgroups)
423 * task_css_check - obtain css for (task, subsys) w/ extra access conds
424 * @task: the target task
425 * @subsys_id: the target subsystem ID
426 * @__c: extra condition expression to be passed to rcu_dereference_check()
428 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
429 * synchronization rules are the same as task_css_set_check().
431 #define task_css_check(task, subsys_id, __c) \
432 task_css_set_check((task), (__c))->subsys[(subsys_id)]
435 * task_css_set - obtain a task's css_set
436 * @task: the task to obtain css_set for
438 * See task_css_set_check().
440 static inline struct css_set *task_css_set(struct task_struct *task)
442 return task_css_set_check(task, false);
446 * task_css - obtain css for (task, subsys)
447 * @task: the target task
448 * @subsys_id: the target subsystem ID
450 * See task_css_check().
452 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
455 return task_css_check(task, subsys_id, false);
459 * task_get_css - find and get the css for (task, subsys)
460 * @task: the target task
461 * @subsys_id: the target subsystem ID
463 * Find the css for the (@task, @subsys_id) combination, increment a
464 * reference on and return it. This function is guaranteed to return a
467 static inline struct cgroup_subsys_state *
468 task_get_css(struct task_struct *task, int subsys_id)
470 struct cgroup_subsys_state *css;
474 css = task_css(task, subsys_id);
475 if (likely(css_tryget_online(css)))
484 * task_css_is_root - test whether a task belongs to the root css
485 * @task: the target task
486 * @subsys_id: the target subsystem ID
488 * Test whether @task belongs to the root css on the specified subsystem.
489 * May be invoked in any context.
491 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
493 return task_css_check(task, subsys_id, true) ==
494 init_css_set.subsys[subsys_id];
497 static inline struct cgroup *task_cgroup(struct task_struct *task,
500 return task_css(task, subsys_id)->cgroup;
504 * cgroup_is_descendant - test ancestry
505 * @cgrp: the cgroup to be tested
506 * @ancestor: possible ancestor of @cgrp
508 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
509 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
510 * and @ancestor are accessible.
512 static inline bool cgroup_is_descendant(struct cgroup *cgrp,
513 struct cgroup *ancestor)
515 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level)
517 return cgrp->ancestor_ids[ancestor->level] == ancestor->id;
521 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry
522 * @task: the task to be tested
523 * @ancestor: possible ancestor of @task's cgroup
525 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
526 * It follows all the same rules as cgroup_is_descendant, and only applies
527 * to the default hierarchy.
529 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
530 struct cgroup *ancestor)
532 struct css_set *cset = task_css_set(task);
534 return cgroup_is_descendant(cset->dfl_cgrp, ancestor);
537 /* no synchronization, the result can only be used as a hint */
538 static inline bool cgroup_is_populated(struct cgroup *cgrp)
540 return cgrp->populated_cnt;
543 /* returns ino associated with a cgroup */
544 static inline ino_t cgroup_ino(struct cgroup *cgrp)
546 return cgrp->kn->ino;
549 /* cft/css accessors for cftype->write() operation */
550 static inline struct cftype *of_cft(struct kernfs_open_file *of)
555 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
557 /* cft/css accessors for cftype->seq_*() operations */
558 static inline struct cftype *seq_cft(struct seq_file *seq)
560 return of_cft(seq->private);
563 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
565 return of_css(seq->private);
569 * Name / path handling functions. All are thin wrappers around the kernfs
570 * counterparts and can be called under any context.
573 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
575 return kernfs_name(cgrp->kn, buf, buflen);
578 static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen)
580 return kernfs_path(cgrp->kn, buf, buflen);
583 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
585 pr_cont_kernfs_name(cgrp->kn);
588 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
590 pr_cont_kernfs_path(cgrp->kn);
593 static inline void cgroup_init_kthreadd(void)
596 * kthreadd is inherited by all kthreads, keep it in the root so
597 * that the new kthreads are guaranteed to stay in the root until
598 * initialization is finished.
600 current->no_cgroup_migration = 1;
603 static inline void cgroup_kthread_ready(void)
606 * This kthread finished initialization. The creator should have
607 * set PF_NO_SETAFFINITY if this kthread should stay in the root.
609 current->no_cgroup_migration = 0;
612 #else /* !CONFIG_CGROUPS */
614 struct cgroup_subsys_state;
617 static inline void css_put(struct cgroup_subsys_state *css) {}
618 static inline int cgroup_attach_task_all(struct task_struct *from,
619 struct task_struct *t) { return 0; }
620 static inline int cgroupstats_build(struct cgroupstats *stats,
621 struct dentry *dentry) { return -EINVAL; }
623 static inline void cgroup_fork(struct task_struct *p) {}
624 static inline int cgroup_can_fork(struct task_struct *p) { return 0; }
625 static inline void cgroup_cancel_fork(struct task_struct *p) {}
626 static inline void cgroup_post_fork(struct task_struct *p) {}
627 static inline void cgroup_exit(struct task_struct *p) {}
628 static inline void cgroup_free(struct task_struct *p) {}
630 static inline int cgroup_init_early(void) { return 0; }
631 static inline int cgroup_init(void) { return 0; }
632 static inline void cgroup_init_kthreadd(void) {}
633 static inline void cgroup_kthread_ready(void) {}
635 static inline bool task_under_cgroup_hierarchy(struct task_struct *task,
636 struct cgroup *ancestor)
640 #endif /* !CONFIG_CGROUPS */
643 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
644 * definition in cgroup-defs.h.
646 #ifdef CONFIG_SOCK_CGROUP_DATA
648 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
649 extern spinlock_t cgroup_sk_update_lock;
652 void cgroup_sk_alloc_disable(void);
653 void cgroup_sk_alloc(struct sock_cgroup_data *skcd);
654 void cgroup_sk_free(struct sock_cgroup_data *skcd);
656 static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
658 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
662 * @skcd->val is 64bit but the following is safe on 32bit too as we
663 * just need the lower ulong to be written and read atomically.
665 v = READ_ONCE(skcd->val);
668 return &cgrp_dfl_root.cgrp;
670 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
672 return (struct cgroup *)(unsigned long)skcd->val;
676 #else /* CONFIG_CGROUP_DATA */
678 static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {}
679 static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {}
681 #endif /* CONFIG_CGROUP_DATA */
683 struct cgroup_namespace {
686 struct user_namespace *user_ns;
687 struct ucounts *ucounts;
688 struct css_set *root_cset;
691 extern struct cgroup_namespace init_cgroup_ns;
693 #ifdef CONFIG_CGROUPS
695 void free_cgroup_ns(struct cgroup_namespace *ns);
697 struct cgroup_namespace *copy_cgroup_ns(unsigned long flags,
698 struct user_namespace *user_ns,
699 struct cgroup_namespace *old_ns);
701 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
702 struct cgroup_namespace *ns);
704 #else /* !CONFIG_CGROUPS */
706 static inline void free_cgroup_ns(struct cgroup_namespace *ns) { }
707 static inline struct cgroup_namespace *
708 copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns,
709 struct cgroup_namespace *old_ns)
714 #endif /* !CONFIG_CGROUPS */
716 static inline void get_cgroup_ns(struct cgroup_namespace *ns)
719 refcount_inc(&ns->count);
722 static inline void put_cgroup_ns(struct cgroup_namespace *ns)
724 if (ns && refcount_dec_and_test(&ns->count))
728 #endif /* _LINUX_CGROUP_H */