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 struct rcu_head rcu_head
;
159 unsigned long flags
; /* "unsigned long" so bitops work */
162 * count users of this cgroup. >0 means busy, but doesn't
163 * necessarily indicate the number of tasks in the cgroup
167 int id
; /* ida allocated in-hierarchy ID */
170 * We link our 'sibling' struct into our parent's 'children'.
171 * Our children link their 'sibling' into our 'children'.
173 struct list_head sibling
; /* my parent's children */
174 struct list_head children
; /* my children */
175 struct list_head files
; /* my files */
177 struct cgroup
*parent
; /* my parent */
178 struct dentry
*dentry
; /* cgroup fs entry, RCU protected */
181 * This is a copy of dentry->d_name, and it's needed because
182 * we can't use dentry->d_name in cgroup_path().
184 * You must acquire rcu_read_lock() to access cgrp->name, and
185 * the only place that can change it is rename(), which is
186 * protected by parent dir's i_mutex.
188 * Normally you should use cgroup_name() wrapper rather than
189 * access it directly.
191 struct cgroup_name __rcu
*name
;
193 /* Private pointers for each registered subsystem */
194 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
196 struct cgroupfs_root
*root
;
197 struct cgroup
*top_cgroup
;
200 * List of cg_cgroup_links pointing at css_sets with
201 * tasks in this cgroup. Protected by css_set_lock
203 struct list_head css_sets
;
205 struct list_head allcg_node
; /* cgroupfs_root->allcg_list */
206 struct list_head cft_q_node
; /* used during cftype add/rm */
209 * Linked list running through all cgroups that can
210 * potentially be reaped by the release agent. Protected by
213 struct list_head release_list
;
216 * list of pidlists, up to two for each namespace (one for procs, one
217 * for tasks); created on demand.
219 struct list_head pidlists
;
220 struct mutex pidlist_mutex
;
222 /* For RCU-protected deletion */
223 struct rcu_head rcu_head
;
224 struct work_struct free_work
;
226 /* List of events which userspace want to receive */
227 struct list_head event_list
;
228 spinlock_t event_list_lock
;
230 /* directory xattrs */
231 struct simple_xattrs xattrs
;
235 * A css_set is a structure holding pointers to a set of
236 * cgroup_subsys_state objects. This saves space in the task struct
237 * object and speeds up fork()/exit(), since a single inc/dec and a
238 * list_add()/del() can bump the reference count on the entire cgroup
244 /* Reference count */
248 * List running through all cgroup groups in the same hash
249 * slot. Protected by css_set_lock
251 struct hlist_node hlist
;
254 * List running through all tasks using this cgroup
255 * group. Protected by css_set_lock
257 struct list_head tasks
;
260 * List of cg_cgroup_link objects on link chains from
261 * cgroups referenced from this css_set. Protected by
264 struct list_head cg_links
;
267 * Set of subsystem states, one for each subsystem. This array
268 * is immutable after creation apart from the init_css_set
269 * during subsystem registration (at boot time) and modular subsystem
272 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
274 /* For RCU-protected deletion */
275 struct rcu_head rcu_head
;
279 * cgroup_map_cb is an abstract callback API for reporting map-valued
283 struct cgroup_map_cb
{
284 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
289 * struct cftype: handler definitions for cgroup control files
291 * When reading/writing to a file:
292 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
293 * - the 'cftype' of the file is file->f_dentry->d_fsdata
297 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
298 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
300 #define MAX_CFTYPE_NAME 64
304 * By convention, the name should begin with the name of the
305 * subsystem, followed by a period. Zero length string indicates
306 * end of cftype array.
308 char name
[MAX_CFTYPE_NAME
];
311 * If not 0, file mode is set to this value, otherwise it will
312 * be figured out automatically
317 * If non-zero, defines the maximum length of string that can
318 * be passed to write_string; defaults to 64
320 size_t max_write_len
;
326 struct simple_xattrs xattrs
;
328 int (*open
)(struct inode
*inode
, struct file
*file
);
329 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
331 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
333 * read_u64() is a shortcut for the common case of returning a
334 * single integer. Use it in place of read()
336 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
338 * read_s64() is a signed version of read_u64()
340 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
342 * read_map() is used for defining a map of key/value
343 * pairs. It should call cb->fill(cb, key, value) for each
344 * entry. The key/value pairs (and their ordering) should not
345 * change between reboots.
347 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
348 struct cgroup_map_cb
*cb
);
350 * read_seq_string() is used for outputting a simple sequence
353 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
356 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
358 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
361 * write_u64() is a shortcut for the common case of accepting
362 * a single integer (as parsed by simple_strtoull) from
363 * userspace. Use in place of write(); return 0 or error.
365 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
367 * write_s64() is a signed version of write_u64()
369 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
372 * write_string() is passed a nul-terminated kernelspace
373 * buffer of maximum length determined by max_write_len.
374 * Returns 0 or -ve error code.
376 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
379 * trigger() callback can be used to get some kick from the
380 * userspace, when the actual string written is not important
381 * at all. The private field can be used to determine the
382 * kick type for multiplexing.
384 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
386 int (*release
)(struct inode
*inode
, struct file
*file
);
389 * register_event() callback will be used to add new userspace
390 * waiter for changes related to the cftype. Implement it if
391 * you want to provide this functionality. Use eventfd_signal()
392 * on eventfd to send notification to userspace.
394 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
395 struct eventfd_ctx
*eventfd
, const char *args
);
397 * unregister_event() callback will be called when userspace
398 * closes the eventfd or on cgroup removing.
399 * This callback must be implemented, if you want provide
400 * notification functionality.
402 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
403 struct eventfd_ctx
*eventfd
);
407 * cftype_sets describe cftypes belonging to a subsystem and are chained at
408 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
409 * terminated by zero length name.
412 struct list_head node
; /* chained at subsys->cftsets */
416 struct cgroup_scanner
{
418 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
419 void (*process_task
)(struct task_struct
*p
,
420 struct cgroup_scanner
*scan
);
421 struct ptr_heap
*heap
;
425 /* Caller should hold rcu_read_lock() */
426 static inline const char *cgroup_name(const struct cgroup
*cgrp
)
428 return rcu_dereference(cgrp
->name
)->name
;
431 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
432 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
434 int cgroup_is_removed(const struct cgroup
*cgrp
);
436 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
438 int cgroup_task_count(const struct cgroup
*cgrp
);
440 /* Return true if cgrp is a descendant of the task's cgroup */
441 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
444 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
447 struct cgroup_taskset
;
448 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
449 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
450 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
451 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
454 * cgroup_taskset_for_each - iterate cgroup_taskset
455 * @task: the loop cursor
456 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
457 * @tset: taskset to iterate
459 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
460 for ((task) = cgroup_taskset_first((tset)); (task); \
461 (task) = cgroup_taskset_next((tset))) \
462 if (!(skip_cgrp) || \
463 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
466 * Control Group subsystem type.
467 * See Documentation/cgroups/cgroups.txt for details
470 struct cgroup_subsys
{
471 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup
*cgrp
);
472 int (*css_online
)(struct cgroup
*cgrp
);
473 void (*css_offline
)(struct cgroup
*cgrp
);
474 void (*css_free
)(struct cgroup
*cgrp
);
476 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
477 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
478 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
479 void (*fork
)(struct task_struct
*task
);
480 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
481 struct task_struct
*task
);
482 void (*bind
)(struct cgroup
*root
);
489 * True if this subsys uses ID. ID is not available before cgroup_init()
490 * (not available in early_init time.)
495 * If %false, this subsystem is properly hierarchical -
496 * configuration, resource accounting and restriction on a parent
497 * cgroup cover those of its children. If %true, hierarchy support
498 * is broken in some ways - some subsystems ignore hierarchy
499 * completely while others are only implemented half-way.
501 * It's now disallowed to create nested cgroups if the subsystem is
502 * broken and cgroup core will emit a warning message on such
503 * cases. Eventually, all subsystems will be made properly
504 * hierarchical and this will go away.
506 bool broken_hierarchy
;
507 bool warned_broken_hierarchy
;
509 #define MAX_CGROUP_TYPE_NAMELEN 32
513 * Link to parent, and list entry in parent's children.
514 * Protected by cgroup_lock()
516 struct cgroupfs_root
*root
;
517 struct list_head sibling
;
518 /* used when use_id == true */
522 /* list of cftype_sets */
523 struct list_head cftsets
;
525 /* base cftypes, automatically [de]registered with subsys itself */
526 struct cftype
*base_cftypes
;
527 struct cftype_set base_cftset
;
529 /* should be defined only by modular subsystems */
530 struct module
*module
;
533 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
534 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
535 #include <linux/cgroup_subsys.h>
536 #undef IS_SUBSYS_ENABLED
539 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
540 struct cgroup
*cgrp
, int subsys_id
)
542 return cgrp
->subsys
[subsys_id
];
546 * function to get the cgroup_subsys_state which allows for extra
547 * rcu_dereference_check() conditions, such as locks used during the
548 * cgroup_subsys::attach() methods.
550 #define task_subsys_state_check(task, subsys_id, __c) \
551 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
552 lockdep_is_held(&task->alloc_lock) || \
553 cgroup_lock_is_held() || (__c))
555 static inline struct cgroup_subsys_state
*
556 task_subsys_state(struct task_struct
*task
, int subsys_id
)
558 return task_subsys_state_check(task
, subsys_id
, false);
561 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
564 return task_subsys_state(task
, subsys_id
)->cgroup
;
568 * cgroup_for_each_child - iterate through children of a cgroup
569 * @pos: the cgroup * to use as the loop cursor
570 * @cgroup: cgroup whose children to walk
572 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
573 * cgroup which hasn't finished ->css_online() or already has finished
574 * ->css_offline() may show up during traversal and it's each subsystem's
575 * responsibility to verify that each @pos is alive.
577 * If a subsystem synchronizes against the parent in its ->css_online() and
578 * before starting iterating, a cgroup which finished ->css_online() is
579 * guaranteed to be visible in the future iterations.
581 #define cgroup_for_each_child(pos, cgroup) \
582 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
584 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
585 struct cgroup
*cgroup
);
586 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
);
589 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
590 * @pos: the cgroup * to use as the loop cursor
591 * @cgroup: cgroup whose descendants to walk
593 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
594 * descendant cgroup which hasn't finished ->css_online() or already has
595 * finished ->css_offline() may show up during traversal and it's each
596 * subsystem's responsibility to verify that each @pos is alive.
598 * If a subsystem synchronizes against the parent in its ->css_online() and
599 * before starting iterating, and synchronizes against @pos on each
600 * iteration, any descendant cgroup which finished ->css_offline() is
601 * guaranteed to be visible in the future iterations.
603 * In other words, the following guarantees that a descendant can't escape
604 * state updates of its ancestors.
608 * Lock @cgrp->parent and @cgrp;
609 * Inherit state from @cgrp->parent;
613 * my_update_state(@cgrp)
616 * Update @cgrp's state;
619 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
621 * Verify @pos is alive and inherit state from @pos->parent;
626 * As long as the inheriting step, including checking the parent state, is
627 * enclosed inside @pos locking, double-locking the parent isn't necessary
628 * while inheriting. The state update to the parent is guaranteed to be
629 * visible by walking order and, as long as inheriting operations to the
630 * same @pos are atomic to each other, multiple updates racing each other
631 * still result in the correct state. It's guaranateed that at least one
632 * inheritance happens for any cgroup after the latest update to its
635 * If checking parent's state requires locking the parent, each inheriting
636 * iteration should lock and unlock both @pos->parent and @pos.
638 * Alternatively, a subsystem may choose to use a single global lock to
639 * synchronize ->css_online() and ->css_offline() against tree-walking
642 #define cgroup_for_each_descendant_pre(pos, cgroup) \
643 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
644 pos = cgroup_next_descendant_pre((pos), (cgroup)))
646 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
647 struct cgroup
*cgroup
);
650 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
651 * @pos: the cgroup * to use as the loop cursor
652 * @cgroup: cgroup whose descendants to walk
654 * Similar to cgroup_for_each_descendant_pre() but performs post-order
655 * traversal instead. Note that the walk visibility guarantee described in
656 * pre-order walk doesn't apply the same to post-order walks.
658 #define cgroup_for_each_descendant_post(pos, cgroup) \
659 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
660 pos = cgroup_next_descendant_post((pos), (cgroup)))
662 /* A cgroup_iter should be treated as an opaque object */
664 struct list_head
*cg_link
;
665 struct list_head
*task
;
669 * To iterate across the tasks in a cgroup:
671 * 1) call cgroup_iter_start to initialize an iterator
673 * 2) call cgroup_iter_next() to retrieve member tasks until it
674 * returns NULL or until you want to end the iteration
676 * 3) call cgroup_iter_end() to destroy the iterator.
678 * Or, call cgroup_scan_tasks() to iterate through every task in a
679 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
680 * the test_task() callback, but not while calling the process_task()
683 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
684 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
685 struct cgroup_iter
*it
);
686 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
687 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
688 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
689 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
692 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
693 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
694 * CSS ID is assigned at cgroup allocation (create) automatically
695 * and removed when subsys calls free_css_id() function. This is because
696 * the lifetime of cgroup_subsys_state is subsys's matter.
698 * Looking up and scanning function should be called under rcu_read_lock().
699 * Taking cgroup_mutex is not necessary for following calls.
700 * But the css returned by this routine can be "not populated yet" or "being
701 * destroyed". The caller should check css and cgroup's status.
705 * Typically Called at ->destroy(), or somewhere the subsys frees
706 * cgroup_subsys_state.
708 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
710 /* Find a cgroup_subsys_state which has given ID */
712 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
715 * Get a cgroup whose id is greater than or equal to id under tree of root.
716 * Returning a cgroup_subsys_state or NULL.
718 struct cgroup_subsys_state
*css_get_next(struct cgroup_subsys
*ss
, int id
,
719 struct cgroup_subsys_state
*root
, int *foundid
);
721 /* Returns true if root is ancestor of cg */
722 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
723 const struct cgroup_subsys_state
*root
);
725 /* Get id and depth of css */
726 unsigned short css_id(struct cgroup_subsys_state
*css
);
727 unsigned short css_depth(struct cgroup_subsys_state
*css
);
728 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
730 #else /* !CONFIG_CGROUPS */
732 static inline int cgroup_init_early(void) { return 0; }
733 static inline int cgroup_init(void) { return 0; }
734 static inline void cgroup_fork(struct task_struct
*p
) {}
735 static inline void cgroup_post_fork(struct task_struct
*p
) {}
736 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
738 static inline void cgroup_lock(void) {}
739 static inline void cgroup_unlock(void) {}
740 static inline int cgroupstats_build(struct cgroupstats
*stats
,
741 struct dentry
*dentry
)
746 /* No cgroups - nothing to do */
747 static inline int cgroup_attach_task_all(struct task_struct
*from
,
748 struct task_struct
*t
)
753 #endif /* !CONFIG_CGROUPS */
755 #endif /* _LINUX_CGROUP_H */