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/rcupdate.h>
15 #include <linux/rculist.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/prio_heap.h>
18 #include <linux/rwsem.h>
19 #include <linux/idr.h>
20 #include <linux/workqueue.h>
21 #include <linux/xattr.h>
31 extern int cgroup_init_early(void);
32 extern int cgroup_init(void);
33 extern void cgroup_lock(void);
34 extern int cgroup_lock_is_held(void);
35 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
36 extern void cgroup_unlock(void);
37 extern void cgroup_fork(struct task_struct *p);
38 extern void cgroup_post_fork(struct task_struct *p);
39 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
40 extern int cgroupstats_build(struct cgroupstats *stats,
41 struct dentry *dentry);
42 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
43 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
45 extern const struct file_operations proc_cgroup_operations;
48 * Define the enumeration of all cgroup subsystems.
50 * We define ids for builtin subsystems and then modular ones.
52 #define SUBSYS(_x) _x ## _subsys_id,
53 enum cgroup_subsys_id {
54 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
55 #include <linux/cgroup_subsys.h>
56 #undef IS_SUBSYS_ENABLED
57 CGROUP_BUILTIN_SUBSYS_COUNT,
59 __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
61 #define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
62 #include <linux/cgroup_subsys.h>
63 #undef IS_SUBSYS_ENABLED
68 /* Per-subsystem/per-cgroup state maintained by the system. */
69 struct cgroup_subsys_state {
71 * The cgroup that this subsystem is attached to. Useful
72 * for subsystems that want to know about the cgroup
75 struct cgroup *cgroup;
78 * State maintained by the cgroup system to allow subsystems
79 * to be "busy". Should be accessed via css_get(),
80 * css_tryget() and css_put().
86 /* ID for this css, if possible */
87 struct css_id __rcu *id;
89 /* Used to put @cgroup->dentry on the last css_put() */
90 struct work_struct dput_work;
93 /* bits in struct cgroup_subsys_state flags field */
95 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
96 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
99 /* Caller must verify that the css is not for root cgroup */
100 static inline void __css_get(struct cgroup_subsys_state *css, int count)
102 atomic_add(count, &css->refcnt);
106 * Call css_get() to hold a reference on the css; it can be used
107 * for a reference obtained via:
108 * - an existing ref-counted reference to the css
109 * - task->cgroups for a locked task
112 static inline void css_get(struct cgroup_subsys_state *css)
114 /* We don't need to reference count the root state */
115 if (!(css->flags & CSS_ROOT))
120 * Call css_tryget() to take a reference on a css if your existing
121 * (known-valid) reference isn't already ref-counted. Returns false if
122 * the css has been destroyed.
125 extern bool __css_tryget(struct cgroup_subsys_state *css);
126 static inline bool css_tryget(struct cgroup_subsys_state *css)
128 if (css->flags & CSS_ROOT)
130 return __css_tryget(css);
134 * css_put() should be called to release a reference taken by
135 * css_get() or css_tryget()
138 extern void __css_put(struct cgroup_subsys_state *css);
139 static inline void css_put(struct cgroup_subsys_state *css)
141 if (!(css->flags & CSS_ROOT))
145 /* bits in struct cgroup flags field */
147 /* Control Group is dead */
150 * Control Group has previously had a child cgroup or a task,
151 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
154 /* Control Group requires release notifications to userspace */
155 CGRP_NOTIFY_ON_RELEASE,
157 * Clone the parent's configuration when creating a new child
158 * cpuset cgroup. For historical reasons, this option can be
159 * specified at mount time and thus is implemented here.
161 CGRP_CPUSET_CLONE_CHILDREN,
165 struct rcu_head rcu_head;
170 unsigned long flags; /* "unsigned long" so bitops work */
173 * count users of this cgroup. >0 means busy, but doesn't
174 * necessarily indicate the number of tasks in the cgroup
178 int id; /* ida allocated in-hierarchy ID */
181 * We link our 'sibling' struct into our parent's 'children'.
182 * Our children link their 'sibling' into our 'children'.
184 struct list_head sibling; /* my parent's children */
185 struct list_head children; /* my children */
186 struct list_head files; /* my files */
188 struct cgroup *parent; /* my parent */
189 struct dentry *dentry; /* cgroup fs entry, RCU protected */
192 * This is a copy of dentry->d_name, and it's needed because
193 * we can't use dentry->d_name in cgroup_path().
195 * You must acquire rcu_read_lock() to access cgrp->name, and
196 * the only place that can change it is rename(), which is
197 * protected by parent dir's i_mutex.
199 * Normally you should use cgroup_name() wrapper rather than
200 * access it directly.
202 struct cgroup_name __rcu *name;
204 /* Private pointers for each registered subsystem */
205 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
207 struct cgroupfs_root *root;
208 struct cgroup *top_cgroup;
211 * List of cg_cgroup_links pointing at css_sets with
212 * tasks in this cgroup. Protected by css_set_lock
214 struct list_head css_sets;
216 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
217 struct list_head cft_q_node; /* used during cftype add/rm */
220 * Linked list running through all cgroups that can
221 * potentially be reaped by the release agent. Protected by
224 struct list_head release_list;
227 * list of pidlists, up to two for each namespace (one for procs, one
228 * for tasks); created on demand.
230 struct list_head pidlists;
231 struct mutex pidlist_mutex;
233 /* For RCU-protected deletion */
234 struct rcu_head rcu_head;
235 struct work_struct free_work;
237 /* List of events which userspace want to receive */
238 struct list_head event_list;
239 spinlock_t event_list_lock;
241 /* directory xattrs */
242 struct simple_xattrs xattrs;
246 * A css_set is a structure holding pointers to a set of
247 * cgroup_subsys_state objects. This saves space in the task struct
248 * object and speeds up fork()/exit(), since a single inc/dec and a
249 * list_add()/del() can bump the reference count on the entire cgroup
255 /* Reference count */
259 * List running through all cgroup groups in the same hash
260 * slot. Protected by css_set_lock
262 struct hlist_node hlist;
265 * List running through all tasks using this cgroup
266 * group. Protected by css_set_lock
268 struct list_head tasks;
271 * List of cg_cgroup_link objects on link chains from
272 * cgroups referenced from this css_set. Protected by
275 struct list_head cg_links;
278 * Set of subsystem states, one for each subsystem. This array
279 * is immutable after creation apart from the init_css_set
280 * during subsystem registration (at boot time) and modular subsystem
283 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
285 /* For RCU-protected deletion */
286 struct rcu_head rcu_head;
290 * cgroup_map_cb is an abstract callback API for reporting map-valued
294 struct cgroup_map_cb {
295 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
300 * struct cftype: handler definitions for cgroup control files
302 * When reading/writing to a file:
303 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
304 * - the 'cftype' of the file is file->f_dentry->d_fsdata
308 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
309 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
311 #define MAX_CFTYPE_NAME 64
315 * By convention, the name should begin with the name of the
316 * subsystem, followed by a period. Zero length string indicates
317 * end of cftype array.
319 char name[MAX_CFTYPE_NAME];
322 * If not 0, file mode is set to this value, otherwise it will
323 * be figured out automatically
328 * If non-zero, defines the maximum length of string that can
329 * be passed to write_string; defaults to 64
331 size_t max_write_len;
337 struct simple_xattrs xattrs;
339 int (*open)(struct inode *inode, struct file *file);
340 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
342 char __user *buf, size_t nbytes, loff_t *ppos);
344 * read_u64() is a shortcut for the common case of returning a
345 * single integer. Use it in place of read()
347 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
349 * read_s64() is a signed version of read_u64()
351 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
353 * read_map() is used for defining a map of key/value
354 * pairs. It should call cb->fill(cb, key, value) for each
355 * entry. The key/value pairs (and their ordering) should not
356 * change between reboots.
358 int (*read_map)(struct cgroup *cont, struct cftype *cft,
359 struct cgroup_map_cb *cb);
361 * read_seq_string() is used for outputting a simple sequence
364 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
367 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
369 const char __user *buf, size_t nbytes, loff_t *ppos);
372 * write_u64() is a shortcut for the common case of accepting
373 * a single integer (as parsed by simple_strtoull) from
374 * userspace. Use in place of write(); return 0 or error.
376 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
378 * write_s64() is a signed version of write_u64()
380 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
383 * write_string() is passed a nul-terminated kernelspace
384 * buffer of maximum length determined by max_write_len.
385 * Returns 0 or -ve error code.
387 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
390 * trigger() callback can be used to get some kick from the
391 * userspace, when the actual string written is not important
392 * at all. The private field can be used to determine the
393 * kick type for multiplexing.
395 int (*trigger)(struct cgroup *cgrp, unsigned int event);
397 int (*release)(struct inode *inode, struct file *file);
400 * register_event() callback will be used to add new userspace
401 * waiter for changes related to the cftype. Implement it if
402 * you want to provide this functionality. Use eventfd_signal()
403 * on eventfd to send notification to userspace.
405 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
406 struct eventfd_ctx *eventfd, const char *args);
408 * unregister_event() callback will be called when userspace
409 * closes the eventfd or on cgroup removing.
410 * This callback must be implemented, if you want provide
411 * notification functionality.
413 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
414 struct eventfd_ctx *eventfd);
418 * cftype_sets describe cftypes belonging to a subsystem and are chained at
419 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
420 * terminated by zero length name.
423 struct list_head node; /* chained at subsys->cftsets */
427 struct cgroup_scanner {
429 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
430 void (*process_task)(struct task_struct *p,
431 struct cgroup_scanner *scan);
432 struct ptr_heap *heap;
436 /* Caller should hold rcu_read_lock() */
437 static inline const char *cgroup_name(const struct cgroup *cgrp)
439 return rcu_dereference(cgrp->name)->name;
442 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
443 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
445 int cgroup_is_removed(const struct cgroup *cgrp);
447 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
449 int cgroup_task_count(const struct cgroup *cgrp);
452 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
455 struct cgroup_taskset;
456 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
457 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
458 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
459 int cgroup_taskset_size(struct cgroup_taskset *tset);
462 * cgroup_taskset_for_each - iterate cgroup_taskset
463 * @task: the loop cursor
464 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
465 * @tset: taskset to iterate
467 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
468 for ((task) = cgroup_taskset_first((tset)); (task); \
469 (task) = cgroup_taskset_next((tset))) \
470 if (!(skip_cgrp) || \
471 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
474 * Control Group subsystem type.
475 * See Documentation/cgroups/cgroups.txt for details
478 struct cgroup_subsys {
479 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
480 int (*css_online)(struct cgroup *cgrp);
481 void (*css_offline)(struct cgroup *cgrp);
482 void (*css_free)(struct cgroup *cgrp);
484 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
485 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
486 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
487 void (*fork)(struct task_struct *task);
488 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
489 struct task_struct *task);
490 void (*bind)(struct cgroup *root);
497 * True if this subsys uses ID. ID is not available before cgroup_init()
498 * (not available in early_init time.)
503 * If %false, this subsystem is properly hierarchical -
504 * configuration, resource accounting and restriction on a parent
505 * cgroup cover those of its children. If %true, hierarchy support
506 * is broken in some ways - some subsystems ignore hierarchy
507 * completely while others are only implemented half-way.
509 * It's now disallowed to create nested cgroups if the subsystem is
510 * broken and cgroup core will emit a warning message on such
511 * cases. Eventually, all subsystems will be made properly
512 * hierarchical and this will go away.
514 bool broken_hierarchy;
515 bool warned_broken_hierarchy;
517 #define MAX_CGROUP_TYPE_NAMELEN 32
521 * Link to parent, and list entry in parent's children.
522 * Protected by cgroup_lock()
524 struct cgroupfs_root *root;
525 struct list_head sibling;
526 /* used when use_id == true */
530 /* list of cftype_sets */
531 struct list_head cftsets;
533 /* base cftypes, automatically [de]registered with subsys itself */
534 struct cftype *base_cftypes;
535 struct cftype_set base_cftset;
537 /* should be defined only by modular subsystems */
538 struct module *module;
541 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
542 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
543 #include <linux/cgroup_subsys.h>
544 #undef IS_SUBSYS_ENABLED
547 static inline struct cgroup_subsys_state *cgroup_subsys_state(
548 struct cgroup *cgrp, int subsys_id)
550 return cgrp->subsys[subsys_id];
554 * function to get the cgroup_subsys_state which allows for extra
555 * rcu_dereference_check() conditions, such as locks used during the
556 * cgroup_subsys::attach() methods.
558 #define task_subsys_state_check(task, subsys_id, __c) \
559 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
560 lockdep_is_held(&task->alloc_lock) || \
561 cgroup_lock_is_held() || (__c))
563 static inline struct cgroup_subsys_state *
564 task_subsys_state(struct task_struct *task, int subsys_id)
566 return task_subsys_state_check(task, subsys_id, false);
569 static inline struct cgroup* task_cgroup(struct task_struct *task,
572 return task_subsys_state(task, subsys_id)->cgroup;
576 * cgroup_for_each_child - iterate through children of a cgroup
577 * @pos: the cgroup * to use as the loop cursor
578 * @cgroup: cgroup whose children to walk
580 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
581 * cgroup which hasn't finished ->css_online() or already has finished
582 * ->css_offline() may show up during traversal and it's each subsystem's
583 * responsibility to verify that each @pos is alive.
585 * If a subsystem synchronizes against the parent in its ->css_online() and
586 * before starting iterating, a cgroup which finished ->css_online() is
587 * guaranteed to be visible in the future iterations.
589 #define cgroup_for_each_child(pos, cgroup) \
590 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
592 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
593 struct cgroup *cgroup);
594 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
597 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
598 * @pos: the cgroup * to use as the loop cursor
599 * @cgroup: cgroup whose descendants to walk
601 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
602 * descendant cgroup which hasn't finished ->css_online() or already has
603 * finished ->css_offline() may show up during traversal and it's each
604 * subsystem's responsibility to verify that each @pos is alive.
606 * If a subsystem synchronizes against the parent in its ->css_online() and
607 * before starting iterating, and synchronizes against @pos on each
608 * iteration, any descendant cgroup which finished ->css_offline() is
609 * guaranteed to be visible in the future iterations.
611 * In other words, the following guarantees that a descendant can't escape
612 * state updates of its ancestors.
616 * Lock @cgrp->parent and @cgrp;
617 * Inherit state from @cgrp->parent;
621 * my_update_state(@cgrp)
624 * Update @cgrp's state;
627 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
629 * Verify @pos is alive and inherit state from @pos->parent;
634 * As long as the inheriting step, including checking the parent state, is
635 * enclosed inside @pos locking, double-locking the parent isn't necessary
636 * while inheriting. The state update to the parent is guaranteed to be
637 * visible by walking order and, as long as inheriting operations to the
638 * same @pos are atomic to each other, multiple updates racing each other
639 * still result in the correct state. It's guaranateed that at least one
640 * inheritance happens for any cgroup after the latest update to its
643 * If checking parent's state requires locking the parent, each inheriting
644 * iteration should lock and unlock both @pos->parent and @pos.
646 * Alternatively, a subsystem may choose to use a single global lock to
647 * synchronize ->css_online() and ->css_offline() against tree-walking
650 #define cgroup_for_each_descendant_pre(pos, cgroup) \
651 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
652 pos = cgroup_next_descendant_pre((pos), (cgroup)))
654 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
655 struct cgroup *cgroup);
658 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
659 * @pos: the cgroup * to use as the loop cursor
660 * @cgroup: cgroup whose descendants to walk
662 * Similar to cgroup_for_each_descendant_pre() but performs post-order
663 * traversal instead. Note that the walk visibility guarantee described in
664 * pre-order walk doesn't apply the same to post-order walks.
666 #define cgroup_for_each_descendant_post(pos, cgroup) \
667 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
668 pos = cgroup_next_descendant_post((pos), (cgroup)))
670 /* A cgroup_iter should be treated as an opaque object */
672 struct list_head *cg_link;
673 struct list_head *task;
677 * To iterate across the tasks in a cgroup:
679 * 1) call cgroup_iter_start to initialize an iterator
681 * 2) call cgroup_iter_next() to retrieve member tasks until it
682 * returns NULL or until you want to end the iteration
684 * 3) call cgroup_iter_end() to destroy the iterator.
686 * Or, call cgroup_scan_tasks() to iterate through every task in a
687 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
688 * the test_task() callback, but not while calling the process_task()
691 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
692 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
693 struct cgroup_iter *it);
694 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
695 int cgroup_scan_tasks(struct cgroup_scanner *scan);
696 int cgroup_attach_task(struct cgroup *, struct task_struct *);
697 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
700 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
701 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
702 * CSS ID is assigned at cgroup allocation (create) automatically
703 * and removed when subsys calls free_css_id() function. This is because
704 * the lifetime of cgroup_subsys_state is subsys's matter.
706 * Looking up and scanning function should be called under rcu_read_lock().
707 * Taking cgroup_mutex is not necessary for following calls.
708 * But the css returned by this routine can be "not populated yet" or "being
709 * destroyed". The caller should check css and cgroup's status.
713 * Typically Called at ->destroy(), or somewhere the subsys frees
714 * cgroup_subsys_state.
716 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
718 /* Find a cgroup_subsys_state which has given ID */
720 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
723 * Get a cgroup whose id is greater than or equal to id under tree of root.
724 * Returning a cgroup_subsys_state or NULL.
726 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
727 struct cgroup_subsys_state *root, int *foundid);
729 /* Returns true if root is ancestor of cg */
730 bool css_is_ancestor(struct cgroup_subsys_state *cg,
731 const struct cgroup_subsys_state *root);
733 /* Get id and depth of css */
734 unsigned short css_id(struct cgroup_subsys_state *css);
735 unsigned short css_depth(struct cgroup_subsys_state *css);
736 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
738 #else /* !CONFIG_CGROUPS */
740 static inline int cgroup_init_early(void) { return 0; }
741 static inline int cgroup_init(void) { return 0; }
742 static inline void cgroup_fork(struct task_struct *p) {}
743 static inline void cgroup_post_fork(struct task_struct *p) {}
744 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
746 static inline void cgroup_lock(void) {}
747 static inline void cgroup_unlock(void) {}
748 static inline int cgroupstats_build(struct cgroupstats *stats,
749 struct dentry *dentry)
754 /* No cgroups - nothing to do */
755 static inline int cgroup_attach_task_all(struct task_struct *from,
756 struct task_struct *t)
761 #endif /* !CONFIG_CGROUPS */
763 #endif /* _LINUX_CGROUP_H */