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
;
47 /* Define the enumeration of all builtin cgroup subsystems */
48 #define SUBSYS(_x) _x ## _subsys_id,
49 #define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
50 enum cgroup_subsys_id
{
51 #include <linux/cgroup_subsys.h>
54 #undef IS_SUBSYS_ENABLED
57 /* Per-subsystem/per-cgroup state maintained by the system. */
58 struct cgroup_subsys_state
{
60 * The cgroup that this subsystem is attached to. Useful
61 * for subsystems that want to know about the cgroup
64 struct cgroup
*cgroup
;
67 * State maintained by the cgroup system to allow subsystems
68 * to be "busy". Should be accessed via css_get(),
69 * css_tryget() and css_put().
75 /* ID for this css, if possible */
76 struct css_id __rcu
*id
;
78 /* Used to put @cgroup->dentry on the last css_put() */
79 struct work_struct dput_work
;
82 /* bits in struct cgroup_subsys_state flags field */
84 CSS_ROOT
= (1 << 0), /* this CSS is the root of the subsystem */
85 CSS_ONLINE
= (1 << 1), /* between ->css_online() and ->css_offline() */
88 /* Caller must verify that the css is not for root cgroup */
89 static inline void __css_get(struct cgroup_subsys_state
*css
, int count
)
91 atomic_add(count
, &css
->refcnt
);
95 * Call css_get() to hold a reference on the css; it can be used
96 * for a reference obtained via:
97 * - an existing ref-counted reference to the css
98 * - task->cgroups for a locked task
101 static inline void css_get(struct cgroup_subsys_state
*css
)
103 /* We don't need to reference count the root state */
104 if (!(css
->flags
& CSS_ROOT
))
109 * Call css_tryget() to take a reference on a css if your existing
110 * (known-valid) reference isn't already ref-counted. Returns false if
111 * the css has been destroyed.
114 extern bool __css_tryget(struct cgroup_subsys_state
*css
);
115 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
117 if (css
->flags
& CSS_ROOT
)
119 return __css_tryget(css
);
123 * css_put() should be called to release a reference taken by
124 * css_get() or css_tryget()
127 extern void __css_put(struct cgroup_subsys_state
*css
);
128 static inline void css_put(struct cgroup_subsys_state
*css
)
130 if (!(css
->flags
& CSS_ROOT
))
134 /* bits in struct cgroup flags field */
136 /* Control Group is dead */
139 * Control Group has previously had a child cgroup or a task,
140 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
143 /* Control Group requires release notifications to userspace */
144 CGRP_NOTIFY_ON_RELEASE
,
146 * Clone the parent's configuration when creating a new child
147 * cpuset cgroup. For historical reasons, this option can be
148 * specified at mount time and thus is implemented here.
150 CGRP_CPUSET_CLONE_CHILDREN
,
154 unsigned long flags
; /* "unsigned long" so bitops work */
157 * count users of this cgroup. >0 means busy, but doesn't
158 * necessarily indicate the number of tasks in the cgroup
162 int id
; /* ida allocated in-hierarchy ID */
165 * We link our 'sibling' struct into our parent's 'children'.
166 * Our children link their 'sibling' into our 'children'.
168 struct list_head sibling
; /* my parent's children */
169 struct list_head children
; /* my children */
170 struct list_head files
; /* my files */
172 struct cgroup
*parent
; /* my parent */
173 struct dentry
*dentry
; /* cgroup fs entry, RCU protected */
175 /* Private pointers for each registered subsystem */
176 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
178 struct cgroupfs_root
*root
;
179 struct cgroup
*top_cgroup
;
182 * List of cg_cgroup_links pointing at css_sets with
183 * tasks in this cgroup. Protected by css_set_lock
185 struct list_head css_sets
;
187 struct list_head allcg_node
; /* cgroupfs_root->allcg_list */
188 struct list_head cft_q_node
; /* used during cftype add/rm */
191 * Linked list running through all cgroups that can
192 * potentially be reaped by the release agent. Protected by
195 struct list_head release_list
;
198 * list of pidlists, up to two for each namespace (one for procs, one
199 * for tasks); created on demand.
201 struct list_head pidlists
;
202 struct mutex pidlist_mutex
;
204 /* For RCU-protected deletion */
205 struct rcu_head rcu_head
;
206 struct work_struct free_work
;
208 /* List of events which userspace want to receive */
209 struct list_head event_list
;
210 spinlock_t event_list_lock
;
212 /* directory xattrs */
213 struct simple_xattrs xattrs
;
217 * A css_set is a structure holding pointers to a set of
218 * cgroup_subsys_state objects. This saves space in the task struct
219 * object and speeds up fork()/exit(), since a single inc/dec and a
220 * list_add()/del() can bump the reference count on the entire cgroup
226 /* Reference count */
230 * List running through all cgroup groups in the same hash
231 * slot. Protected by css_set_lock
233 struct hlist_node hlist
;
236 * List running through all tasks using this cgroup
237 * group. Protected by css_set_lock
239 struct list_head tasks
;
242 * List of cg_cgroup_link objects on link chains from
243 * cgroups referenced from this css_set. Protected by
246 struct list_head cg_links
;
249 * Set of subsystem states, one for each subsystem. This array
250 * is immutable after creation apart from the init_css_set
251 * during subsystem registration (at boot time) and modular subsystem
254 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
256 /* For RCU-protected deletion */
257 struct rcu_head rcu_head
;
261 * cgroup_map_cb is an abstract callback API for reporting map-valued
265 struct cgroup_map_cb
{
266 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
271 * struct cftype: handler definitions for cgroup control files
273 * When reading/writing to a file:
274 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
275 * - the 'cftype' of the file is file->f_dentry->d_fsdata
279 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
280 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
282 #define MAX_CFTYPE_NAME 64
286 * By convention, the name should begin with the name of the
287 * subsystem, followed by a period. Zero length string indicates
288 * end of cftype array.
290 char name
[MAX_CFTYPE_NAME
];
293 * If not 0, file mode is set to this value, otherwise it will
294 * be figured out automatically
299 * If non-zero, defines the maximum length of string that can
300 * be passed to write_string; defaults to 64
302 size_t max_write_len
;
308 struct simple_xattrs xattrs
;
310 int (*open
)(struct inode
*inode
, struct file
*file
);
311 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
313 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
315 * read_u64() is a shortcut for the common case of returning a
316 * single integer. Use it in place of read()
318 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
320 * read_s64() is a signed version of read_u64()
322 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
324 * read_map() is used for defining a map of key/value
325 * pairs. It should call cb->fill(cb, key, value) for each
326 * entry. The key/value pairs (and their ordering) should not
327 * change between reboots.
329 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
330 struct cgroup_map_cb
*cb
);
332 * read_seq_string() is used for outputting a simple sequence
335 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
338 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
340 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
343 * write_u64() is a shortcut for the common case of accepting
344 * a single integer (as parsed by simple_strtoull) from
345 * userspace. Use in place of write(); return 0 or error.
347 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
349 * write_s64() is a signed version of write_u64()
351 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
354 * write_string() is passed a nul-terminated kernelspace
355 * buffer of maximum length determined by max_write_len.
356 * Returns 0 or -ve error code.
358 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
361 * trigger() callback can be used to get some kick from the
362 * userspace, when the actual string written is not important
363 * at all. The private field can be used to determine the
364 * kick type for multiplexing.
366 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
368 int (*release
)(struct inode
*inode
, struct file
*file
);
371 * register_event() callback will be used to add new userspace
372 * waiter for changes related to the cftype. Implement it if
373 * you want to provide this functionality. Use eventfd_signal()
374 * on eventfd to send notification to userspace.
376 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
377 struct eventfd_ctx
*eventfd
, const char *args
);
379 * unregister_event() callback will be called when userspace
380 * closes the eventfd or on cgroup removing.
381 * This callback must be implemented, if you want provide
382 * notification functionality.
384 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
385 struct eventfd_ctx
*eventfd
);
389 * cftype_sets describe cftypes belonging to a subsystem and are chained at
390 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
391 * terminated by zero length name.
394 struct list_head node
; /* chained at subsys->cftsets */
398 struct cgroup_scanner
{
400 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
401 void (*process_task
)(struct task_struct
*p
,
402 struct cgroup_scanner
*scan
);
403 struct ptr_heap
*heap
;
407 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
408 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
410 int cgroup_is_removed(const struct cgroup
*cgrp
);
412 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
414 int cgroup_task_count(const struct cgroup
*cgrp
);
416 /* Return true if cgrp is a descendant of the task's cgroup */
417 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
420 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
423 struct cgroup_taskset
;
424 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
425 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
426 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
427 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
430 * cgroup_taskset_for_each - iterate cgroup_taskset
431 * @task: the loop cursor
432 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
433 * @tset: taskset to iterate
435 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
436 for ((task) = cgroup_taskset_first((tset)); (task); \
437 (task) = cgroup_taskset_next((tset))) \
438 if (!(skip_cgrp) || \
439 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
442 * Control Group subsystem type.
443 * See Documentation/cgroups/cgroups.txt for details
446 struct cgroup_subsys
{
447 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup
*cgrp
);
448 int (*css_online
)(struct cgroup
*cgrp
);
449 void (*css_offline
)(struct cgroup
*cgrp
);
450 void (*css_free
)(struct cgroup
*cgrp
);
452 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
453 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
454 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
455 void (*fork
)(struct task_struct
*task
);
456 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
457 struct task_struct
*task
);
458 void (*bind
)(struct cgroup
*root
);
465 * True if this subsys uses ID. ID is not available before cgroup_init()
466 * (not available in early_init time.)
471 * If %false, this subsystem is properly hierarchical -
472 * configuration, resource accounting and restriction on a parent
473 * cgroup cover those of its children. If %true, hierarchy support
474 * is broken in some ways - some subsystems ignore hierarchy
475 * completely while others are only implemented half-way.
477 * It's now disallowed to create nested cgroups if the subsystem is
478 * broken and cgroup core will emit a warning message on such
479 * cases. Eventually, all subsystems will be made properly
480 * hierarchical and this will go away.
482 bool broken_hierarchy
;
483 bool warned_broken_hierarchy
;
485 #define MAX_CGROUP_TYPE_NAMELEN 32
489 * Link to parent, and list entry in parent's children.
490 * Protected by cgroup_lock()
492 struct cgroupfs_root
*root
;
493 struct list_head sibling
;
494 /* used when use_id == true */
498 /* list of cftype_sets */
499 struct list_head cftsets
;
501 /* base cftypes, automatically [de]registered with subsys itself */
502 struct cftype
*base_cftypes
;
503 struct cftype_set base_cftset
;
505 /* should be defined only by modular subsystems */
506 struct module
*module
;
509 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
510 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
511 #include <linux/cgroup_subsys.h>
512 #undef IS_SUBSYS_ENABLED
515 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
516 struct cgroup
*cgrp
, int subsys_id
)
518 return cgrp
->subsys
[subsys_id
];
522 * function to get the cgroup_subsys_state which allows for extra
523 * rcu_dereference_check() conditions, such as locks used during the
524 * cgroup_subsys::attach() methods.
526 #define task_subsys_state_check(task, subsys_id, __c) \
527 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
528 lockdep_is_held(&task->alloc_lock) || \
529 cgroup_lock_is_held() || (__c))
531 static inline struct cgroup_subsys_state
*
532 task_subsys_state(struct task_struct
*task
, int subsys_id
)
534 return task_subsys_state_check(task
, subsys_id
, false);
537 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
540 return task_subsys_state(task
, subsys_id
)->cgroup
;
544 * cgroup_for_each_child - iterate through children of a cgroup
545 * @pos: the cgroup * to use as the loop cursor
546 * @cgroup: cgroup whose children to walk
548 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
549 * cgroup which hasn't finished ->css_online() or already has finished
550 * ->css_offline() may show up during traversal and it's each subsystem's
551 * responsibility to verify that each @pos is alive.
553 * If a subsystem synchronizes against the parent in its ->css_online() and
554 * before starting iterating, a cgroup which finished ->css_online() is
555 * guaranteed to be visible in the future iterations.
557 #define cgroup_for_each_child(pos, cgroup) \
558 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
560 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
561 struct cgroup
*cgroup
);
562 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
);
565 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
566 * @pos: the cgroup * to use as the loop cursor
567 * @cgroup: cgroup whose descendants to walk
569 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
570 * descendant cgroup which hasn't finished ->css_online() or already has
571 * finished ->css_offline() may show up during traversal and it's each
572 * subsystem's responsibility to verify that each @pos is alive.
574 * If a subsystem synchronizes against the parent in its ->css_online() and
575 * before starting iterating, and synchronizes against @pos on each
576 * iteration, any descendant cgroup which finished ->css_offline() is
577 * guaranteed to be visible in the future iterations.
579 * In other words, the following guarantees that a descendant can't escape
580 * state updates of its ancestors.
584 * Lock @cgrp->parent and @cgrp;
585 * Inherit state from @cgrp->parent;
589 * my_update_state(@cgrp)
592 * Update @cgrp's state;
595 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
597 * Verify @pos is alive and inherit state from @pos->parent;
602 * As long as the inheriting step, including checking the parent state, is
603 * enclosed inside @pos locking, double-locking the parent isn't necessary
604 * while inheriting. The state update to the parent is guaranteed to be
605 * visible by walking order and, as long as inheriting operations to the
606 * same @pos are atomic to each other, multiple updates racing each other
607 * still result in the correct state. It's guaranateed that at least one
608 * inheritance happens for any cgroup after the latest update to its
611 * If checking parent's state requires locking the parent, each inheriting
612 * iteration should lock and unlock both @pos->parent and @pos.
614 * Alternatively, a subsystem may choose to use a single global lock to
615 * synchronize ->css_online() and ->css_offline() against tree-walking
618 #define cgroup_for_each_descendant_pre(pos, cgroup) \
619 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
620 pos = cgroup_next_descendant_pre((pos), (cgroup)))
622 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
623 struct cgroup
*cgroup
);
626 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
627 * @pos: the cgroup * to use as the loop cursor
628 * @cgroup: cgroup whose descendants to walk
630 * Similar to cgroup_for_each_descendant_pre() but performs post-order
631 * traversal instead. Note that the walk visibility guarantee described in
632 * pre-order walk doesn't apply the same to post-order walks.
634 #define cgroup_for_each_descendant_post(pos, cgroup) \
635 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
636 pos = cgroup_next_descendant_post((pos), (cgroup)))
638 /* A cgroup_iter should be treated as an opaque object */
640 struct list_head
*cg_link
;
641 struct list_head
*task
;
645 * To iterate across the tasks in a cgroup:
647 * 1) call cgroup_iter_start to initialize an iterator
649 * 2) call cgroup_iter_next() to retrieve member tasks until it
650 * returns NULL or until you want to end the iteration
652 * 3) call cgroup_iter_end() to destroy the iterator.
654 * Or, call cgroup_scan_tasks() to iterate through every task in a
655 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
656 * the test_task() callback, but not while calling the process_task()
659 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
660 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
661 struct cgroup_iter
*it
);
662 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
663 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
664 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
665 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
668 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
669 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
670 * CSS ID is assigned at cgroup allocation (create) automatically
671 * and removed when subsys calls free_css_id() function. This is because
672 * the lifetime of cgroup_subsys_state is subsys's matter.
674 * Looking up and scanning function should be called under rcu_read_lock().
675 * Taking cgroup_mutex is not necessary for following calls.
676 * But the css returned by this routine can be "not populated yet" or "being
677 * destroyed". The caller should check css and cgroup's status.
681 * Typically Called at ->destroy(), or somewhere the subsys frees
682 * cgroup_subsys_state.
684 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
686 /* Find a cgroup_subsys_state which has given ID */
688 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
690 /* Returns true if root is ancestor of cg */
691 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
692 const struct cgroup_subsys_state
*root
);
694 /* Get id and depth of css */
695 unsigned short css_id(struct cgroup_subsys_state
*css
);
696 unsigned short css_depth(struct cgroup_subsys_state
*css
);
697 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
699 #else /* !CONFIG_CGROUPS */
701 static inline int cgroup_init_early(void) { return 0; }
702 static inline int cgroup_init(void) { return 0; }
703 static inline void cgroup_fork(struct task_struct
*p
) {}
704 static inline void cgroup_post_fork(struct task_struct
*p
) {}
705 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
707 static inline void cgroup_lock(void) {}
708 static inline void cgroup_unlock(void) {}
709 static inline int cgroupstats_build(struct cgroupstats
*stats
,
710 struct dentry
*dentry
)
715 /* No cgroups - nothing to do */
716 static inline int cgroup_attach_task_all(struct task_struct
*from
,
717 struct task_struct
*t
)
722 #endif /* !CONFIG_CGROUPS */
724 #endif /* _LINUX_CGROUP_H */