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>
32 extern int cgroup_init_early(void);
33 extern int cgroup_init(void);
34 extern void cgroup_lock(void);
35 extern int cgroup_lock_is_held(void);
36 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
37 extern void cgroup_unlock(void);
38 extern void cgroup_fork(struct task_struct *p);
39 extern void cgroup_post_fork(struct task_struct *p);
40 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
41 extern int cgroupstats_build(struct cgroupstats *stats,
42 struct dentry *dentry);
43 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
44 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
46 extern const struct file_operations proc_cgroup_operations;
48 /* Define the enumeration of all builtin cgroup subsystems */
49 #define SUBSYS(_x) _x ## _subsys_id,
50 #define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
51 enum cgroup_subsys_id {
52 #include <linux/cgroup_subsys.h>
55 #undef IS_SUBSYS_ENABLED
58 /* Per-subsystem/per-cgroup state maintained by the system. */
59 struct cgroup_subsys_state {
61 * The cgroup that this subsystem is attached to. Useful
62 * for subsystems that want to know about the cgroup
65 struct cgroup *cgroup;
68 * State maintained by the cgroup system to allow subsystems
69 * to be "busy". Should be accessed via css_get(),
70 * css_tryget() and css_put().
76 /* ID for this css, if possible */
77 struct css_id __rcu *id;
79 /* Used to put @cgroup->dentry on the last css_put() */
80 struct work_struct dput_work;
83 /* bits in struct cgroup_subsys_state flags field */
85 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
86 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
89 /* Caller must verify that the css is not for root cgroup */
90 static inline void __css_get(struct cgroup_subsys_state *css, int count)
92 atomic_add(count, &css->refcnt);
96 * Call css_get() to hold a reference on the css; it can be used
97 * for a reference obtained via:
98 * - an existing ref-counted reference to the css
99 * - task->cgroups for a locked task
102 static inline void css_get(struct cgroup_subsys_state *css)
104 /* We don't need to reference count the root state */
105 if (!(css->flags & CSS_ROOT))
110 * Call css_tryget() to take a reference on a css if your existing
111 * (known-valid) reference isn't already ref-counted. Returns false if
112 * the css has been destroyed.
115 extern bool __css_tryget(struct cgroup_subsys_state *css);
116 static inline bool css_tryget(struct cgroup_subsys_state *css)
118 if (css->flags & CSS_ROOT)
120 return __css_tryget(css);
124 * css_put() should be called to release a reference taken by
125 * css_get() or css_tryget()
128 extern void __css_put(struct cgroup_subsys_state *css);
129 static inline void css_put(struct cgroup_subsys_state *css)
131 if (!(css->flags & CSS_ROOT))
135 /* bits in struct cgroup flags field */
137 /* Control Group is dead */
140 * Control Group has previously had a child cgroup or a task,
141 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
144 /* Control Group requires release notifications to userspace */
145 CGRP_NOTIFY_ON_RELEASE,
147 * Clone the parent's configuration when creating a new child
148 * cpuset cgroup. For historical reasons, this option can be
149 * specified at mount time and thus is implemented here.
151 CGRP_CPUSET_CLONE_CHILDREN,
155 unsigned long flags; /* "unsigned long" so bitops work */
158 * count users of this cgroup. >0 means busy, but doesn't
159 * necessarily indicate the number of tasks in the cgroup
163 int id; /* ida allocated in-hierarchy ID */
166 * We link our 'sibling' struct into our parent's 'children'.
167 * Our children link their 'sibling' into our 'children'.
169 struct list_head sibling; /* my parent's children */
170 struct list_head children; /* my children */
171 struct list_head files; /* my files */
173 struct cgroup *parent; /* my parent */
174 struct dentry *dentry; /* cgroup fs entry, RCU protected */
176 /* Private pointers for each registered subsystem */
177 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
179 struct cgroupfs_root *root;
180 struct cgroup *top_cgroup;
183 * List of cg_cgroup_links pointing at css_sets with
184 * tasks in this cgroup. Protected by css_set_lock
186 struct list_head css_sets;
188 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
189 struct list_head cft_q_node; /* used during cftype add/rm */
192 * Linked list running through all cgroups that can
193 * potentially be reaped by the release agent. Protected by
196 struct list_head release_list;
199 * list of pidlists, up to two for each namespace (one for procs, one
200 * for tasks); created on demand.
202 struct list_head pidlists;
203 struct mutex pidlist_mutex;
205 /* For RCU-protected deletion */
206 struct rcu_head rcu_head;
207 struct work_struct free_work;
209 /* List of events which userspace want to receive */
210 struct list_head event_list;
211 spinlock_t event_list_lock;
213 /* directory xattrs */
214 struct simple_xattrs xattrs;
218 * A css_set is a structure holding pointers to a set of
219 * cgroup_subsys_state objects. This saves space in the task struct
220 * object and speeds up fork()/exit(), since a single inc/dec and a
221 * list_add()/del() can bump the reference count on the entire cgroup
227 /* Reference count */
231 * List running through all cgroup groups in the same hash
232 * slot. Protected by css_set_lock
234 struct hlist_node hlist;
237 * List running through all tasks using this cgroup
238 * group. Protected by css_set_lock
240 struct list_head tasks;
243 * List of cg_cgroup_link objects on link chains from
244 * cgroups referenced from this css_set. Protected by
247 struct list_head cg_links;
250 * Set of subsystem states, one for each subsystem. This array
251 * is immutable after creation apart from the init_css_set
252 * during subsystem registration (at boot time) and modular subsystem
255 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
257 /* For RCU-protected deletion */
258 struct rcu_head rcu_head;
262 * cgroup_map_cb is an abstract callback API for reporting map-valued
266 struct cgroup_map_cb {
267 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
272 * struct cftype: handler definitions for cgroup control files
274 * When reading/writing to a file:
275 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
276 * - the 'cftype' of the file is file->f_dentry->d_fsdata
280 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
281 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
283 #define MAX_CFTYPE_NAME 64
287 * By convention, the name should begin with the name of the
288 * subsystem, followed by a period. Zero length string indicates
289 * end of cftype array.
291 char name[MAX_CFTYPE_NAME];
294 * If not 0, file mode is set to this value, otherwise it will
295 * be figured out automatically
300 * If non-zero, defines the maximum length of string that can
301 * be passed to write_string; defaults to 64
303 size_t max_write_len;
309 struct simple_xattrs xattrs;
311 int (*open)(struct inode *inode, struct file *file);
312 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
314 char __user *buf, size_t nbytes, loff_t *ppos);
316 * read_u64() is a shortcut for the common case of returning a
317 * single integer. Use it in place of read()
319 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
321 * read_s64() is a signed version of read_u64()
323 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
325 * read_map() is used for defining a map of key/value
326 * pairs. It should call cb->fill(cb, key, value) for each
327 * entry. The key/value pairs (and their ordering) should not
328 * change between reboots.
330 int (*read_map)(struct cgroup *cont, struct cftype *cft,
331 struct cgroup_map_cb *cb);
333 * read_seq_string() is used for outputting a simple sequence
336 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
339 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
341 const char __user *buf, size_t nbytes, loff_t *ppos);
344 * write_u64() is a shortcut for the common case of accepting
345 * a single integer (as parsed by simple_strtoull) from
346 * userspace. Use in place of write(); return 0 or error.
348 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
350 * write_s64() is a signed version of write_u64()
352 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
355 * write_string() is passed a nul-terminated kernelspace
356 * buffer of maximum length determined by max_write_len.
357 * Returns 0 or -ve error code.
359 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
362 * trigger() callback can be used to get some kick from the
363 * userspace, when the actual string written is not important
364 * at all. The private field can be used to determine the
365 * kick type for multiplexing.
367 int (*trigger)(struct cgroup *cgrp, unsigned int event);
369 int (*release)(struct inode *inode, struct file *file);
372 * register_event() callback will be used to add new userspace
373 * waiter for changes related to the cftype. Implement it if
374 * you want to provide this functionality. Use eventfd_signal()
375 * on eventfd to send notification to userspace.
377 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
378 struct eventfd_ctx *eventfd, const char *args);
380 * unregister_event() callback will be called when userspace
381 * closes the eventfd or on cgroup removing.
382 * This callback must be implemented, if you want provide
383 * notification functionality.
385 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
386 struct eventfd_ctx *eventfd);
390 * cftype_sets describe cftypes belonging to a subsystem and are chained at
391 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
392 * terminated by zero length name.
395 struct list_head node; /* chained at subsys->cftsets */
399 struct cgroup_scanner {
401 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
402 void (*process_task)(struct task_struct *p,
403 struct cgroup_scanner *scan);
404 struct ptr_heap *heap;
408 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
409 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
411 int cgroup_is_removed(const struct cgroup *cgrp);
413 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
415 int cgroup_task_count(const struct cgroup *cgrp);
417 /* Return true if cgrp is a descendant of the task's cgroup */
418 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
421 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
424 struct cgroup_taskset;
425 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
426 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
427 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
428 int cgroup_taskset_size(struct cgroup_taskset *tset);
431 * cgroup_taskset_for_each - iterate cgroup_taskset
432 * @task: the loop cursor
433 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
434 * @tset: taskset to iterate
436 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
437 for ((task) = cgroup_taskset_first((tset)); (task); \
438 (task) = cgroup_taskset_next((tset))) \
439 if (!(skip_cgrp) || \
440 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
443 * Control Group subsystem type.
444 * See Documentation/cgroups/cgroups.txt for details
447 struct cgroup_subsys {
448 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
449 int (*css_online)(struct cgroup *cgrp);
450 void (*css_offline)(struct cgroup *cgrp);
451 void (*css_free)(struct cgroup *cgrp);
453 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
454 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
455 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
456 void (*fork)(struct task_struct *task);
457 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
458 struct task_struct *task);
459 void (*bind)(struct cgroup *root);
466 * True if this subsys uses ID. ID is not available before cgroup_init()
467 * (not available in early_init time.)
472 * If %false, this subsystem is properly hierarchical -
473 * configuration, resource accounting and restriction on a parent
474 * cgroup cover those of its children. If %true, hierarchy support
475 * is broken in some ways - some subsystems ignore hierarchy
476 * completely while others are only implemented half-way.
478 * It's now disallowed to create nested cgroups if the subsystem is
479 * broken and cgroup core will emit a warning message on such
480 * cases. Eventually, all subsystems will be made properly
481 * hierarchical and this will go away.
483 bool broken_hierarchy;
484 bool warned_broken_hierarchy;
486 #define MAX_CGROUP_TYPE_NAMELEN 32
490 * Link to parent, and list entry in parent's children.
491 * Protected by cgroup_lock()
493 struct cgroupfs_root *root;
494 struct list_head sibling;
495 /* used when use_id == true */
499 /* list of cftype_sets */
500 struct list_head cftsets;
502 /* base cftypes, automatically [de]registered with subsys itself */
503 struct cftype *base_cftypes;
504 struct cftype_set base_cftset;
506 /* should be defined only by modular subsystems */
507 struct module *module;
510 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
511 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
512 #include <linux/cgroup_subsys.h>
513 #undef IS_SUBSYS_ENABLED
516 static inline struct cgroup_subsys_state *cgroup_subsys_state(
517 struct cgroup *cgrp, int subsys_id)
519 return cgrp->subsys[subsys_id];
523 * function to get the cgroup_subsys_state which allows for extra
524 * rcu_dereference_check() conditions, such as locks used during the
525 * cgroup_subsys::attach() methods.
527 #define task_subsys_state_check(task, subsys_id, __c) \
528 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
529 lockdep_is_held(&task->alloc_lock) || \
530 cgroup_lock_is_held() || (__c))
532 static inline struct cgroup_subsys_state *
533 task_subsys_state(struct task_struct *task, int subsys_id)
535 return task_subsys_state_check(task, subsys_id, false);
538 static inline struct cgroup* task_cgroup(struct task_struct *task,
541 return task_subsys_state(task, subsys_id)->cgroup;
545 * cgroup_for_each_child - iterate through children of a cgroup
546 * @pos: the cgroup * to use as the loop cursor
547 * @cgroup: cgroup whose children to walk
549 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
550 * cgroup which hasn't finished ->css_online() or already has finished
551 * ->css_offline() may show up during traversal and it's each subsystem's
552 * responsibility to verify that each @pos is alive.
554 * If a subsystem synchronizes against the parent in its ->css_online() and
555 * before starting iterating, a cgroup which finished ->css_online() is
556 * guaranteed to be visible in the future iterations.
558 #define cgroup_for_each_child(pos, cgroup) \
559 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
561 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
562 struct cgroup *cgroup);
563 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
566 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
567 * @pos: the cgroup * to use as the loop cursor
568 * @cgroup: cgroup whose descendants to walk
570 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
571 * descendant cgroup which hasn't finished ->css_online() or already has
572 * finished ->css_offline() may show up during traversal and it's each
573 * subsystem's responsibility to verify that each @pos is alive.
575 * If a subsystem synchronizes against the parent in its ->css_online() and
576 * before starting iterating, and synchronizes against @pos on each
577 * iteration, any descendant cgroup which finished ->css_offline() is
578 * guaranteed to be visible in the future iterations.
580 * In other words, the following guarantees that a descendant can't escape
581 * state updates of its ancestors.
585 * Lock @cgrp->parent and @cgrp;
586 * Inherit state from @cgrp->parent;
590 * my_update_state(@cgrp)
593 * Update @cgrp's state;
596 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
598 * Verify @pos is alive and inherit state from @pos->parent;
603 * As long as the inheriting step, including checking the parent state, is
604 * enclosed inside @pos locking, double-locking the parent isn't necessary
605 * while inheriting. The state update to the parent is guaranteed to be
606 * visible by walking order and, as long as inheriting operations to the
607 * same @pos are atomic to each other, multiple updates racing each other
608 * still result in the correct state. It's guaranateed that at least one
609 * inheritance happens for any cgroup after the latest update to its
612 * If checking parent's state requires locking the parent, each inheriting
613 * iteration should lock and unlock both @pos->parent and @pos.
615 * Alternatively, a subsystem may choose to use a single global lock to
616 * synchronize ->css_online() and ->css_offline() against tree-walking
619 #define cgroup_for_each_descendant_pre(pos, cgroup) \
620 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
621 pos = cgroup_next_descendant_pre((pos), (cgroup)))
623 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
624 struct cgroup *cgroup);
627 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
628 * @pos: the cgroup * to use as the loop cursor
629 * @cgroup: cgroup whose descendants to walk
631 * Similar to cgroup_for_each_descendant_pre() but performs post-order
632 * traversal instead. Note that the walk visibility guarantee described in
633 * pre-order walk doesn't apply the same to post-order walks.
635 #define cgroup_for_each_descendant_post(pos, cgroup) \
636 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
637 pos = cgroup_next_descendant_post((pos), (cgroup)))
639 /* A cgroup_iter should be treated as an opaque object */
641 struct list_head *cg_link;
642 struct list_head *task;
646 * To iterate across the tasks in a cgroup:
648 * 1) call cgroup_iter_start to initialize an iterator
650 * 2) call cgroup_iter_next() to retrieve member tasks until it
651 * returns NULL or until you want to end the iteration
653 * 3) call cgroup_iter_end() to destroy the iterator.
655 * Or, call cgroup_scan_tasks() to iterate through every task in a
656 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
657 * the test_task() callback, but not while calling the process_task()
660 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
661 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
662 struct cgroup_iter *it);
663 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
664 int cgroup_scan_tasks(struct cgroup_scanner *scan);
665 int cgroup_attach_task(struct cgroup *, struct task_struct *);
666 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
669 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
670 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
671 * CSS ID is assigned at cgroup allocation (create) automatically
672 * and removed when subsys calls free_css_id() function. This is because
673 * the lifetime of cgroup_subsys_state is subsys's matter.
675 * Looking up and scanning function should be called under rcu_read_lock().
676 * Taking cgroup_mutex is not necessary for following calls.
677 * But the css returned by this routine can be "not populated yet" or "being
678 * destroyed". The caller should check css and cgroup's status.
682 * Typically Called at ->destroy(), or somewhere the subsys frees
683 * cgroup_subsys_state.
685 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
687 /* Find a cgroup_subsys_state which has given ID */
689 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
692 * Get a cgroup whose id is greater than or equal to id under tree of root.
693 * Returning a cgroup_subsys_state or NULL.
695 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
696 struct cgroup_subsys_state *root, int *foundid);
698 /* Returns true if root is ancestor of cg */
699 bool css_is_ancestor(struct cgroup_subsys_state *cg,
700 const struct cgroup_subsys_state *root);
702 /* Get id and depth of css */
703 unsigned short css_id(struct cgroup_subsys_state *css);
704 unsigned short css_depth(struct cgroup_subsys_state *css);
705 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
707 #else /* !CONFIG_CGROUPS */
709 static inline int cgroup_init_early(void) { return 0; }
710 static inline int cgroup_init(void) { return 0; }
711 static inline void cgroup_fork(struct task_struct *p) {}
712 static inline void cgroup_post_fork(struct task_struct *p) {}
713 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
715 static inline void cgroup_lock(void) {}
716 static inline void cgroup_unlock(void) {}
717 static inline int cgroupstats_build(struct cgroupstats *stats,
718 struct dentry *dentry)
723 /* No cgroups - nothing to do */
724 static inline int cgroup_attach_task_all(struct task_struct *from,
725 struct task_struct *t)
730 #endif /* !CONFIG_CGROUPS */
732 #endif /* _LINUX_CGROUP_H */