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 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 ->post_create() and ->pre_destroy() */
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 cgroup values when creating a new child cgroup
152 unsigned long flags
; /* "unsigned long" so bitops work */
155 * count users of this cgroup. >0 means busy, but doesn't
156 * necessarily indicate the number of tasks in the cgroup
161 * We link our 'sibling' struct into our parent's 'children'.
162 * Our children link their 'sibling' into our 'children'.
164 struct list_head sibling
; /* my parent's children */
165 struct list_head children
; /* my children */
166 struct list_head files
; /* my files */
168 struct cgroup
*parent
; /* my parent */
169 struct dentry
*dentry
; /* cgroup fs entry, RCU protected */
171 /* Private pointers for each registered subsystem */
172 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
174 struct cgroupfs_root
*root
;
175 struct cgroup
*top_cgroup
;
178 * List of cg_cgroup_links pointing at css_sets with
179 * tasks in this cgroup. Protected by css_set_lock
181 struct list_head css_sets
;
183 struct list_head allcg_node
; /* cgroupfs_root->allcg_list */
184 struct list_head cft_q_node
; /* used during cftype add/rm */
187 * Linked list running through all cgroups that can
188 * potentially be reaped by the release agent. Protected by
191 struct list_head release_list
;
194 * list of pidlists, up to two for each namespace (one for procs, one
195 * for tasks); created on demand.
197 struct list_head pidlists
;
198 struct mutex pidlist_mutex
;
200 /* For RCU-protected deletion */
201 struct rcu_head rcu_head
;
203 /* List of events which userspace want to receive */
204 struct list_head event_list
;
205 spinlock_t event_list_lock
;
207 /* directory xattrs */
208 struct simple_xattrs xattrs
;
212 * A css_set is a structure holding pointers to a set of
213 * cgroup_subsys_state objects. This saves space in the task struct
214 * object and speeds up fork()/exit(), since a single inc/dec and a
215 * list_add()/del() can bump the reference count on the entire cgroup
221 /* Reference count */
225 * List running through all cgroup groups in the same hash
226 * slot. Protected by css_set_lock
228 struct hlist_node hlist
;
231 * List running through all tasks using this cgroup
232 * group. Protected by css_set_lock
234 struct list_head tasks
;
237 * List of cg_cgroup_link objects on link chains from
238 * cgroups referenced from this css_set. Protected by
241 struct list_head cg_links
;
244 * Set of subsystem states, one for each subsystem. This array
245 * is immutable after creation apart from the init_css_set
246 * during subsystem registration (at boot time) and modular subsystem
249 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
251 /* For RCU-protected deletion */
252 struct rcu_head rcu_head
;
256 * cgroup_map_cb is an abstract callback API for reporting map-valued
260 struct cgroup_map_cb
{
261 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
266 * struct cftype: handler definitions for cgroup control files
268 * When reading/writing to a file:
269 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
270 * - the 'cftype' of the file is file->f_dentry->d_fsdata
274 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
275 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
277 #define MAX_CFTYPE_NAME 64
281 * By convention, the name should begin with the name of the
282 * subsystem, followed by a period. Zero length string indicates
283 * end of cftype array.
285 char name
[MAX_CFTYPE_NAME
];
288 * If not 0, file mode is set to this value, otherwise it will
289 * be figured out automatically
294 * If non-zero, defines the maximum length of string that can
295 * be passed to write_string; defaults to 64
297 size_t max_write_len
;
303 struct simple_xattrs xattrs
;
305 int (*open
)(struct inode
*inode
, struct file
*file
);
306 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
308 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
310 * read_u64() is a shortcut for the common case of returning a
311 * single integer. Use it in place of read()
313 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
315 * read_s64() is a signed version of read_u64()
317 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
319 * read_map() is used for defining a map of key/value
320 * pairs. It should call cb->fill(cb, key, value) for each
321 * entry. The key/value pairs (and their ordering) should not
322 * change between reboots.
324 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
325 struct cgroup_map_cb
*cb
);
327 * read_seq_string() is used for outputting a simple sequence
330 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
333 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
335 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
338 * write_u64() is a shortcut for the common case of accepting
339 * a single integer (as parsed by simple_strtoull) from
340 * userspace. Use in place of write(); return 0 or error.
342 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
344 * write_s64() is a signed version of write_u64()
346 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
349 * write_string() is passed a nul-terminated kernelspace
350 * buffer of maximum length determined by max_write_len.
351 * Returns 0 or -ve error code.
353 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
356 * trigger() callback can be used to get some kick from the
357 * userspace, when the actual string written is not important
358 * at all. The private field can be used to determine the
359 * kick type for multiplexing.
361 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
363 int (*release
)(struct inode
*inode
, struct file
*file
);
366 * register_event() callback will be used to add new userspace
367 * waiter for changes related to the cftype. Implement it if
368 * you want to provide this functionality. Use eventfd_signal()
369 * on eventfd to send notification to userspace.
371 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
372 struct eventfd_ctx
*eventfd
, const char *args
);
374 * unregister_event() callback will be called when userspace
375 * closes the eventfd or on cgroup removing.
376 * This callback must be implemented, if you want provide
377 * notification functionality.
379 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
380 struct eventfd_ctx
*eventfd
);
384 * cftype_sets describe cftypes belonging to a subsystem and are chained at
385 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
386 * terminated by zero length name.
389 struct list_head node
; /* chained at subsys->cftsets */
393 struct cgroup_scanner
{
395 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
396 void (*process_task
)(struct task_struct
*p
,
397 struct cgroup_scanner
*scan
);
398 struct ptr_heap
*heap
;
402 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
403 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
405 int cgroup_is_removed(const struct cgroup
*cgrp
);
407 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
409 int cgroup_task_count(const struct cgroup
*cgrp
);
411 /* Return true if cgrp is a descendant of the task's cgroup */
412 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
415 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
418 struct cgroup_taskset
;
419 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
420 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
421 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
422 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
425 * cgroup_taskset_for_each - iterate cgroup_taskset
426 * @task: the loop cursor
427 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
428 * @tset: taskset to iterate
430 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
431 for ((task) = cgroup_taskset_first((tset)); (task); \
432 (task) = cgroup_taskset_next((tset))) \
433 if (!(skip_cgrp) || \
434 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
437 * Control Group subsystem type.
438 * See Documentation/cgroups/cgroups.txt for details
441 struct cgroup_subsys
{
442 struct cgroup_subsys_state
*(*create
)(struct cgroup
*cgrp
);
443 int (*post_create
)(struct cgroup
*cgrp
);
444 void (*pre_destroy
)(struct cgroup
*cgrp
);
445 void (*destroy
)(struct cgroup
*cgrp
);
446 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
447 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
448 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
449 void (*fork
)(struct task_struct
*task
);
450 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
451 struct task_struct
*task
);
452 void (*post_clone
)(struct cgroup
*cgrp
);
453 void (*bind
)(struct cgroup
*root
);
460 * True if this subsys uses ID. ID is not available before cgroup_init()
461 * (not available in early_init time.)
466 * If %false, this subsystem is properly hierarchical -
467 * configuration, resource accounting and restriction on a parent
468 * cgroup cover those of its children. If %true, hierarchy support
469 * is broken in some ways - some subsystems ignore hierarchy
470 * completely while others are only implemented half-way.
472 * It's now disallowed to create nested cgroups if the subsystem is
473 * broken and cgroup core will emit a warning message on such
474 * cases. Eventually, all subsystems will be made properly
475 * hierarchical and this will go away.
477 bool broken_hierarchy
;
478 bool warned_broken_hierarchy
;
480 #define MAX_CGROUP_TYPE_NAMELEN 32
484 * Link to parent, and list entry in parent's children.
485 * Protected by cgroup_lock()
487 struct cgroupfs_root
*root
;
488 struct list_head sibling
;
489 /* used when use_id == true */
493 /* list of cftype_sets */
494 struct list_head cftsets
;
496 /* base cftypes, automatically [de]registered with subsys itself */
497 struct cftype
*base_cftypes
;
498 struct cftype_set base_cftset
;
500 /* should be defined only by modular subsystems */
501 struct module
*module
;
504 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
505 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
506 #include <linux/cgroup_subsys.h>
507 #undef IS_SUBSYS_ENABLED
510 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
511 struct cgroup
*cgrp
, int subsys_id
)
513 return cgrp
->subsys
[subsys_id
];
517 * function to get the cgroup_subsys_state which allows for extra
518 * rcu_dereference_check() conditions, such as locks used during the
519 * cgroup_subsys::attach() methods.
521 #define task_subsys_state_check(task, subsys_id, __c) \
522 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
523 lockdep_is_held(&task->alloc_lock) || \
524 cgroup_lock_is_held() || (__c))
526 static inline struct cgroup_subsys_state
*
527 task_subsys_state(struct task_struct
*task
, int subsys_id
)
529 return task_subsys_state_check(task
, subsys_id
, false);
532 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
535 return task_subsys_state(task
, subsys_id
)->cgroup
;
539 * cgroup_for_each_child - iterate through children of a cgroup
540 * @pos: the cgroup * to use as the loop cursor
541 * @cgroup: cgroup whose children to walk
543 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
544 * cgroup which hasn't finished ->post_create() or already has finished
545 * ->pre_destroy() may show up during traversal and it's each subsystem's
546 * responsibility to verify that each @pos is alive.
548 * If a subsystem synchronizes against the parent in its ->post_create()
549 * and before starting iterating, a cgroup which finished ->post_create()
550 * is guaranteed to be visible in the future iterations.
552 #define cgroup_for_each_child(pos, cgroup) \
553 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
555 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
556 struct cgroup
*cgroup
);
559 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
560 * @pos: the cgroup * to use as the loop cursor
561 * @cgroup: cgroup whose descendants to walk
563 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
564 * descendant cgroup which hasn't finished ->post_create() or already has
565 * finished ->pre_destroy() may show up during traversal and it's each
566 * subsystem's responsibility to verify that each @pos is alive.
568 * If a subsystem synchronizes against the parent in its ->post_create()
569 * and before starting iterating, and synchronizes against @pos on each
570 * iteration, any descendant cgroup which finished ->post_create() is
571 * guaranteed to be visible in the future iterations.
573 * In other words, the following guarantees that a descendant can't escape
574 * state updates of its ancestors.
576 * my_post_create(@cgrp)
578 * Lock @cgrp->parent and @cgrp;
579 * Inherit state from @cgrp->parent;
583 * my_update_state(@cgrp)
586 * Update @cgrp's state;
589 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
591 * Verify @pos is alive and inherit state from @pos->parent;
596 * As long as the inheriting step, including checking the parent state, is
597 * enclosed inside @pos locking, double-locking the parent isn't necessary
598 * while inheriting. The state update to the parent is guaranteed to be
599 * visible by walking order and, as long as inheriting operations to the
600 * same @pos are atomic to each other, multiple updates racing each other
601 * still result in the correct state. It's guaranateed that at least one
602 * inheritance happens for any cgroup after the latest update to its
605 * If checking parent's state requires locking the parent, each inheriting
606 * iteration should lock and unlock both @pos->parent and @pos.
608 * Alternatively, a subsystem may choose to use a single global lock to
609 * synchronize ->post_create() and ->pre_destroy() against tree-walking
612 #define cgroup_for_each_descendant_pre(pos, cgroup) \
613 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
614 pos = cgroup_next_descendant_pre((pos), (cgroup)))
616 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
617 struct cgroup
*cgroup
);
620 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
621 * @pos: the cgroup * to use as the loop cursor
622 * @cgroup: cgroup whose descendants to walk
624 * Similar to cgroup_for_each_descendant_pre() but performs post-order
625 * traversal instead. Note that the walk visibility guarantee described in
626 * pre-order walk doesn't apply the same to post-order walks.
628 #define cgroup_for_each_descendant_post(pos, cgroup) \
629 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
630 pos = cgroup_next_descendant_post((pos), (cgroup)))
632 /* A cgroup_iter should be treated as an opaque object */
634 struct list_head
*cg_link
;
635 struct list_head
*task
;
639 * To iterate across the tasks in a cgroup:
641 * 1) call cgroup_iter_start to initialize an iterator
643 * 2) call cgroup_iter_next() to retrieve member tasks until it
644 * returns NULL or until you want to end the iteration
646 * 3) call cgroup_iter_end() to destroy the iterator.
648 * Or, call cgroup_scan_tasks() to iterate through every task in a
649 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
650 * the test_task() callback, but not while calling the process_task()
653 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
654 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
655 struct cgroup_iter
*it
);
656 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
657 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
658 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
659 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
662 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
663 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
664 * CSS ID is assigned at cgroup allocation (create) automatically
665 * and removed when subsys calls free_css_id() function. This is because
666 * the lifetime of cgroup_subsys_state is subsys's matter.
668 * Looking up and scanning function should be called under rcu_read_lock().
669 * Taking cgroup_mutex is not necessary for following calls.
670 * But the css returned by this routine can be "not populated yet" or "being
671 * destroyed". The caller should check css and cgroup's status.
675 * Typically Called at ->destroy(), or somewhere the subsys frees
676 * cgroup_subsys_state.
678 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
680 /* Find a cgroup_subsys_state which has given ID */
682 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
685 * Get a cgroup whose id is greater than or equal to id under tree of root.
686 * Returning a cgroup_subsys_state or NULL.
688 struct cgroup_subsys_state
*css_get_next(struct cgroup_subsys
*ss
, int id
,
689 struct cgroup_subsys_state
*root
, int *foundid
);
691 /* Returns true if root is ancestor of cg */
692 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
693 const struct cgroup_subsys_state
*root
);
695 /* Get id and depth of css */
696 unsigned short css_id(struct cgroup_subsys_state
*css
);
697 unsigned short css_depth(struct cgroup_subsys_state
*css
);
698 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
700 #else /* !CONFIG_CGROUPS */
702 static inline int cgroup_init_early(void) { return 0; }
703 static inline int cgroup_init(void) { return 0; }
704 static inline void cgroup_fork(struct task_struct
*p
) {}
705 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
706 static inline void cgroup_post_fork(struct task_struct
*p
) {}
707 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
709 static inline void cgroup_lock(void) {}
710 static inline void cgroup_unlock(void) {}
711 static inline int cgroupstats_build(struct cgroupstats
*stats
,
712 struct dentry
*dentry
)
717 /* No cgroups - nothing to do */
718 static inline int cgroup_attach_task_all(struct task_struct
*from
,
719 struct task_struct
*t
)
724 #endif /* !CONFIG_CGROUPS */
726 #endif /* _LINUX_CGROUP_H */