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>
16 #include <linux/rwsem.h>
18 #include <linux/seq_file.h>
19 #include <linux/kernfs.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;
46 extern struct cgroup_root cgrp_dfl_root;
47 extern struct css_set init_css_set;
49 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
50 #include <linux/cgroup_subsys.h>
53 bool css_has_online_children(struct cgroup_subsys_state *css);
54 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
55 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup,
56 struct cgroup_subsys *ss);
57 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
58 struct cgroup_subsys *ss);
60 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
61 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
62 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
64 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
65 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
66 int cgroup_rm_cftypes(struct cftype *cfts);
68 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
69 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry);
70 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
71 struct pid *pid, struct task_struct *tsk);
73 void cgroup_fork(struct task_struct *p);
74 extern int cgroup_can_fork(struct task_struct *p,
75 void *ss_priv[CGROUP_CANFORK_COUNT]);
76 extern void cgroup_cancel_fork(struct task_struct *p,
77 void *ss_priv[CGROUP_CANFORK_COUNT]);
78 extern void cgroup_post_fork(struct task_struct *p,
79 void *old_ss_priv[CGROUP_CANFORK_COUNT]);
80 void cgroup_exit(struct task_struct *p);
82 int cgroup_init_early(void);
83 int cgroup_init(void);
86 * Iteration helpers and macros.
89 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
90 struct cgroup_subsys_state *parent);
91 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos,
92 struct cgroup_subsys_state *css);
93 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos);
94 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos,
95 struct cgroup_subsys_state *css);
97 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
98 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
100 void css_task_iter_start(struct cgroup_subsys_state *css,
101 struct css_task_iter *it);
102 struct task_struct *css_task_iter_next(struct css_task_iter *it);
103 void css_task_iter_end(struct css_task_iter *it);
106 * css_for_each_child - iterate through children of a css
107 * @pos: the css * to use as the loop cursor
108 * @parent: css whose children to walk
110 * Walk @parent's children. Must be called under rcu_read_lock().
112 * If a subsystem synchronizes ->css_online() and the start of iteration, a
113 * css which finished ->css_online() is guaranteed to be visible in the
114 * future iterations and will stay visible until the last reference is put.
115 * A css which hasn't finished ->css_online() or already finished
116 * ->css_offline() may show up during traversal. It's each subsystem's
117 * responsibility to synchronize against on/offlining.
119 * It is allowed to temporarily drop RCU read lock during iteration. The
120 * caller is responsible for ensuring that @pos remains accessible until
121 * the start of the next iteration by, for example, bumping the css refcnt.
123 #define css_for_each_child(pos, parent) \
124 for ((pos) = css_next_child(NULL, (parent)); (pos); \
125 (pos) = css_next_child((pos), (parent)))
128 * css_for_each_descendant_pre - pre-order walk of a css's descendants
129 * @pos: the css * to use as the loop cursor
130 * @root: css whose descendants to walk
132 * Walk @root's descendants. @root is included in the iteration and the
133 * first node to be visited. Must be called under rcu_read_lock().
135 * If a subsystem synchronizes ->css_online() and the start of iteration, a
136 * css which finished ->css_online() is guaranteed to be visible in the
137 * future iterations and will stay visible until the last reference is put.
138 * A css which hasn't finished ->css_online() or already finished
139 * ->css_offline() may show up during traversal. It's each subsystem's
140 * responsibility to synchronize against on/offlining.
142 * For example, the following guarantees that a descendant can't escape
143 * state updates of its ancestors.
147 * Lock @css's parent and @css;
148 * Inherit state from the parent;
152 * my_update_state(@css)
154 * css_for_each_descendant_pre(@pos, @css) {
157 * Update @css's state;
159 * Verify @pos is alive and inherit state from its parent;
164 * As long as the inheriting step, including checking the parent state, is
165 * enclosed inside @pos locking, double-locking the parent isn't necessary
166 * while inheriting. The state update to the parent is guaranteed to be
167 * visible by walking order and, as long as inheriting operations to the
168 * same @pos are atomic to each other, multiple updates racing each other
169 * still result in the correct state. It's guaranateed that at least one
170 * inheritance happens for any css after the latest update to its parent.
172 * If checking parent's state requires locking the parent, each inheriting
173 * iteration should lock and unlock both @pos->parent and @pos.
175 * Alternatively, a subsystem may choose to use a single global lock to
176 * synchronize ->css_online() and ->css_offline() against tree-walking
179 * It is allowed to temporarily drop RCU read lock during iteration. The
180 * caller is responsible for ensuring that @pos remains accessible until
181 * the start of the next iteration by, for example, bumping the css refcnt.
183 #define css_for_each_descendant_pre(pos, css) \
184 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
185 (pos) = css_next_descendant_pre((pos), (css)))
188 * css_for_each_descendant_post - post-order walk of a css's descendants
189 * @pos: the css * to use as the loop cursor
190 * @css: css whose descendants to walk
192 * Similar to css_for_each_descendant_pre() but performs post-order
193 * traversal instead. @root is included in the iteration and the last
194 * node to be visited.
196 * If a subsystem synchronizes ->css_online() and the start of iteration, a
197 * css which finished ->css_online() is guaranteed to be visible in the
198 * future iterations and will stay visible until the last reference is put.
199 * A css which hasn't finished ->css_online() or already finished
200 * ->css_offline() may show up during traversal. It's each subsystem's
201 * responsibility to synchronize against on/offlining.
203 * Note that the walk visibility guarantee example described in pre-order
204 * walk doesn't apply the same to post-order walks.
206 #define css_for_each_descendant_post(pos, css) \
207 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
208 (pos) = css_next_descendant_post((pos), (css)))
211 * cgroup_taskset_for_each - iterate cgroup_taskset
212 * @task: the loop cursor
213 * @tset: taskset to iterate
215 #define cgroup_taskset_for_each(task, tset) \
216 for ((task) = cgroup_taskset_first((tset)); (task); \
217 (task) = cgroup_taskset_next((tset)))
224 * css_get - obtain a reference on the specified css
227 * The caller must already have a reference.
229 static inline void css_get(struct cgroup_subsys_state *css)
231 if (!(css->flags & CSS_NO_REF))
232 percpu_ref_get(&css->refcnt);
236 * css_get_many - obtain references on the specified css
238 * @n: number of references to get
240 * The caller must already have a reference.
242 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n)
244 if (!(css->flags & CSS_NO_REF))
245 percpu_ref_get_many(&css->refcnt, n);
249 * css_tryget - try to obtain a reference on the specified css
252 * Obtain a reference on @css unless it already has reached zero and is
253 * being released. This function doesn't care whether @css is on or
254 * offline. The caller naturally needs to ensure that @css is accessible
255 * but doesn't have to be holding a reference on it - IOW, RCU protected
256 * access is good enough for this function. Returns %true if a reference
257 * count was successfully obtained; %false otherwise.
259 static inline bool css_tryget(struct cgroup_subsys_state *css)
261 if (!(css->flags & CSS_NO_REF))
262 return percpu_ref_tryget(&css->refcnt);
267 * css_tryget_online - try to obtain a reference on the specified css if online
270 * Obtain a reference on @css if it's online. The caller naturally needs
271 * to ensure that @css is accessible but doesn't have to be holding a
272 * reference on it - IOW, RCU protected access is good enough for this
273 * function. Returns %true if a reference count was successfully obtained;
276 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
278 if (!(css->flags & CSS_NO_REF))
279 return percpu_ref_tryget_live(&css->refcnt);
284 * css_put - put a css reference
287 * Put a reference obtained via css_get() and css_tryget_online().
289 static inline void css_put(struct cgroup_subsys_state *css)
291 if (!(css->flags & CSS_NO_REF))
292 percpu_ref_put(&css->refcnt);
296 * css_put_many - put css references
298 * @n: number of references to put
300 * Put references obtained via css_get() and css_tryget_online().
302 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n)
304 if (!(css->flags & CSS_NO_REF))
305 percpu_ref_put_many(&css->refcnt, n);
309 * task_css_set_check - obtain a task's css_set with extra access conditions
310 * @task: the task to obtain css_set for
311 * @__c: extra condition expression to be passed to rcu_dereference_check()
313 * A task's css_set is RCU protected, initialized and exited while holding
314 * task_lock(), and can only be modified while holding both cgroup_mutex
315 * and task_lock() while the task is alive. This macro verifies that the
316 * caller is inside proper critical section and returns @task's css_set.
318 * The caller can also specify additional allowed conditions via @__c, such
319 * as locks used during the cgroup_subsys::attach() methods.
321 #ifdef CONFIG_PROVE_RCU
322 extern struct mutex cgroup_mutex;
323 extern struct rw_semaphore css_set_rwsem;
324 #define task_css_set_check(task, __c) \
325 rcu_dereference_check((task)->cgroups, \
326 lockdep_is_held(&cgroup_mutex) || \
327 lockdep_is_held(&css_set_rwsem) || \
328 ((task)->flags & PF_EXITING) || (__c))
330 #define task_css_set_check(task, __c) \
331 rcu_dereference((task)->cgroups)
335 * task_css_check - obtain css for (task, subsys) w/ extra access conds
336 * @task: the target task
337 * @subsys_id: the target subsystem ID
338 * @__c: extra condition expression to be passed to rcu_dereference_check()
340 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
341 * synchronization rules are the same as task_css_set_check().
343 #define task_css_check(task, subsys_id, __c) \
344 task_css_set_check((task), (__c))->subsys[(subsys_id)]
347 * task_css_set - obtain a task's css_set
348 * @task: the task to obtain css_set for
350 * See task_css_set_check().
352 static inline struct css_set *task_css_set(struct task_struct *task)
354 return task_css_set_check(task, false);
358 * task_css - obtain css for (task, subsys)
359 * @task: the target task
360 * @subsys_id: the target subsystem ID
362 * See task_css_check().
364 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
367 return task_css_check(task, subsys_id, false);
371 * task_get_css - find and get the css for (task, subsys)
372 * @task: the target task
373 * @subsys_id: the target subsystem ID
375 * Find the css for the (@task, @subsys_id) combination, increment a
376 * reference on and return it. This function is guaranteed to return a
379 static inline struct cgroup_subsys_state *
380 task_get_css(struct task_struct *task, int subsys_id)
382 struct cgroup_subsys_state *css;
386 css = task_css(task, subsys_id);
387 if (likely(css_tryget_online(css)))
396 * task_css_is_root - test whether a task belongs to the root css
397 * @task: the target task
398 * @subsys_id: the target subsystem ID
400 * Test whether @task belongs to the root css on the specified subsystem.
401 * May be invoked in any context.
403 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
405 return task_css_check(task, subsys_id, true) ==
406 init_css_set.subsys[subsys_id];
409 static inline struct cgroup *task_cgroup(struct task_struct *task,
412 return task_css(task, subsys_id)->cgroup;
416 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
417 * @cgrp: the cgroup of interest
419 * The default hierarchy is the v2 interface of cgroup and this function
420 * can be used to test whether a cgroup is on the default hierarchy for
421 * cases where a subsystem should behave differnetly depending on the
424 * The set of behaviors which change on the default hierarchy are still
425 * being determined and the mount option is prefixed with __DEVEL__.
427 * List of changed behaviors:
429 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
430 * and "name" are disallowed.
432 * - When mounting an existing superblock, mount options should match.
434 * - Remount is disallowed.
436 * - rename(2) is disallowed.
438 * - "tasks" is removed. Everything should be at process granularity. Use
439 * "cgroup.procs" instead.
441 * - "cgroup.procs" is not sorted. pids will be unique unless they got
442 * recycled inbetween reads.
444 * - "release_agent" and "notify_on_release" are removed. Replacement
445 * notification mechanism will be implemented.
447 * - "cgroup.clone_children" is removed.
449 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
450 * and its descendants contain no task; otherwise, 1. The file also
451 * generates kernfs notification which can be monitored through poll and
452 * [di]notify when the value of the file changes.
454 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
455 * take masks of ancestors with non-empty cpus/mems, instead of being
456 * moved to an ancestor.
458 * - cpuset: a task can be moved into an empty cpuset, and again it takes
459 * masks of ancestors.
461 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
464 * - blkcg: blk-throttle becomes properly hierarchical.
466 * - debug: disallowed on the default hierarchy.
468 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
470 return cgrp->root == &cgrp_dfl_root;
473 /* no synchronization, the result can only be used as a hint */
474 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
476 return !list_empty(&cgrp->cset_links);
479 /* returns ino associated with a cgroup */
480 static inline ino_t cgroup_ino(struct cgroup *cgrp)
482 return cgrp->kn->ino;
485 /* cft/css accessors for cftype->write() operation */
486 static inline struct cftype *of_cft(struct kernfs_open_file *of)
491 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
493 /* cft/css accessors for cftype->seq_*() operations */
494 static inline struct cftype *seq_cft(struct seq_file *seq)
496 return of_cft(seq->private);
499 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
501 return of_css(seq->private);
505 * Name / path handling functions. All are thin wrappers around the kernfs
506 * counterparts and can be called under any context.
509 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
511 return kernfs_name(cgrp->kn, buf, buflen);
514 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
517 return kernfs_path(cgrp->kn, buf, buflen);
520 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
522 pr_cont_kernfs_name(cgrp->kn);
525 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
527 pr_cont_kernfs_path(cgrp->kn);
530 #else /* !CONFIG_CGROUPS */
532 struct cgroup_subsys_state;
534 static inline void css_put(struct cgroup_subsys_state *css) {}
535 static inline int cgroup_attach_task_all(struct task_struct *from,
536 struct task_struct *t) { return 0; }
537 static inline int cgroupstats_build(struct cgroupstats *stats,
538 struct dentry *dentry) { return -EINVAL; }
540 static inline void cgroup_fork(struct task_struct *p) {}
541 static inline int cgroup_can_fork(struct task_struct *p,
542 void *ss_priv[CGROUP_CANFORK_COUNT])
544 static inline void cgroup_cancel_fork(struct task_struct *p,
545 void *ss_priv[CGROUP_CANFORK_COUNT]) {}
546 static inline void cgroup_post_fork(struct task_struct *p,
547 void *ss_priv[CGROUP_CANFORK_COUNT]) {}
548 static inline void cgroup_exit(struct task_struct *p) {}
550 static inline int cgroup_init_early(void) { return 0; }
551 static inline int cgroup_init(void) { return 0; }
553 #endif /* !CONFIG_CGROUPS */
555 #endif /* _LINUX_CGROUP_H */