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/cgroupstats.h>
16 #include <linux/prio_heap.h>
17 #include <linux/rwsem.h>
18 #include <linux/idr.h>
19 #include <linux/workqueue.h>
29 extern int cgroup_init_early(void);
30 extern int cgroup_init(void);
31 extern void cgroup_lock(void);
32 extern int cgroup_lock_is_held(void);
33 extern bool cgroup_lock_live_group(struct cgroup
*cgrp
);
34 extern void cgroup_unlock(void);
35 extern void cgroup_fork(struct task_struct
*p
);
36 extern void cgroup_fork_callbacks(struct task_struct
*p
);
37 extern void cgroup_post_fork(struct task_struct
*p
);
38 extern void cgroup_exit(struct task_struct
*p
, int run_callbacks
);
39 extern int cgroupstats_build(struct cgroupstats
*stats
,
40 struct dentry
*dentry
);
41 extern int cgroup_load_subsys(struct cgroup_subsys
*ss
);
42 extern void cgroup_unload_subsys(struct cgroup_subsys
*ss
);
44 extern const struct file_operations proc_cgroup_operations
;
46 /* Define the enumeration of all builtin cgroup subsystems */
47 #define SUBSYS(_x) _x ## _subsys_id,
48 enum cgroup_subsys_id
{
49 #include <linux/cgroup_subsys.h>
50 CGROUP_BUILTIN_SUBSYS_COUNT
54 * This define indicates the maximum number of subsystems that can be loaded
55 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
56 * track of all of them.
58 #define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
60 /* Per-subsystem/per-cgroup state maintained by the system. */
61 struct cgroup_subsys_state
{
63 * The cgroup that this subsystem is attached to. Useful
64 * for subsystems that want to know about the cgroup
67 struct cgroup
*cgroup
;
70 * State maintained by the cgroup system to allow subsystems
71 * to be "busy". Should be accessed via css_get(),
72 * css_tryget() and and css_put().
78 /* ID for this css, if possible */
79 struct css_id __rcu
*id
;
81 /* Used to put @cgroup->dentry on the last css_put() */
82 struct work_struct dput_work
;
85 /* bits in struct cgroup_subsys_state flags field */
87 CSS_ROOT
, /* This CSS is the root of the subsystem */
88 CSS_REMOVED
, /* This CSS is dead */
89 CSS_CLEAR_CSS_REFS
, /* @ss->__DEPRECATED_clear_css_refs */
92 /* Caller must verify that the css is not for root cgroup */
93 static inline void __css_get(struct cgroup_subsys_state
*css
, int count
)
95 atomic_add(count
, &css
->refcnt
);
99 * Call css_get() to hold a reference on the css; it can be used
100 * for a reference obtained via:
101 * - an existing ref-counted reference to the css
102 * - task->cgroups for a locked task
105 static inline void css_get(struct cgroup_subsys_state
*css
)
107 /* We don't need to reference count the root state */
108 if (!test_bit(CSS_ROOT
, &css
->flags
))
112 static inline bool css_is_removed(struct cgroup_subsys_state
*css
)
114 return test_bit(CSS_REMOVED
, &css
->flags
);
118 * Call css_tryget() to take a reference on a css if your existing
119 * (known-valid) reference isn't already ref-counted. Returns false if
120 * the css has been destroyed.
123 extern bool __css_tryget(struct cgroup_subsys_state
*css
);
124 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
126 if (test_bit(CSS_ROOT
, &css
->flags
))
128 return __css_tryget(css
);
132 * css_put() should be called to release a reference taken by
133 * css_get() or css_tryget()
136 extern void __css_put(struct cgroup_subsys_state
*css
);
137 static inline void css_put(struct cgroup_subsys_state
*css
)
139 if (!test_bit(CSS_ROOT
, &css
->flags
))
143 /* bits in struct cgroup flags field */
145 /* Control Group is dead */
148 * Control Group has previously had a child cgroup or a task,
149 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
152 /* Control Group requires release notifications to userspace */
153 CGRP_NOTIFY_ON_RELEASE
,
155 * A thread in rmdir() is wating for this cgroup.
159 * Clone cgroup values when creating a new child cgroup
165 unsigned long flags
; /* "unsigned long" so bitops work */
168 * count users of this cgroup. >0 means busy, but doesn't
169 * necessarily indicate the number of tasks in the cgroup
174 * We link our 'sibling' struct into our parent's 'children'.
175 * Our children link their 'sibling' into our 'children'.
177 struct list_head sibling
; /* my parent's children */
178 struct list_head children
; /* my children */
179 struct list_head files
; /* my files */
181 struct cgroup
*parent
; /* my parent */
182 struct dentry __rcu
*dentry
; /* cgroup fs entry, RCU protected */
184 /* Private pointers for each registered subsystem */
185 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
187 struct cgroupfs_root
*root
;
188 struct cgroup
*top_cgroup
;
191 * List of cg_cgroup_links pointing at css_sets with
192 * tasks in this cgroup. Protected by css_set_lock
194 struct list_head css_sets
;
196 struct list_head allcg_node
; /* cgroupfs_root->allcg_list */
197 struct list_head cft_q_node
; /* used during cftype add/rm */
200 * Linked list running through all cgroups that can
201 * potentially be reaped by the release agent. Protected by
204 struct list_head release_list
;
207 * list of pidlists, up to two for each namespace (one for procs, one
208 * for tasks); created on demand.
210 struct list_head pidlists
;
211 struct mutex pidlist_mutex
;
213 /* For RCU-protected deletion */
214 struct rcu_head rcu_head
;
216 /* List of events which userspace want to receive */
217 struct list_head event_list
;
218 spinlock_t event_list_lock
;
222 * A css_set is a structure holding pointers to a set of
223 * cgroup_subsys_state objects. This saves space in the task struct
224 * object and speeds up fork()/exit(), since a single inc/dec and a
225 * list_add()/del() can bump the reference count on the entire cgroup
231 /* Reference count */
235 * List running through all cgroup groups in the same hash
236 * slot. Protected by css_set_lock
238 struct hlist_node hlist
;
241 * List running through all tasks using this cgroup
242 * group. Protected by css_set_lock
244 struct list_head tasks
;
247 * List of cg_cgroup_link objects on link chains from
248 * cgroups referenced from this css_set. Protected by
251 struct list_head cg_links
;
254 * Set of subsystem states, one for each subsystem. This array
255 * is immutable after creation apart from the init_css_set
256 * during subsystem registration (at boot time) and modular subsystem
259 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
261 /* For RCU-protected deletion */
262 struct rcu_head rcu_head
;
266 * cgroup_map_cb is an abstract callback API for reporting map-valued
270 struct cgroup_map_cb
{
271 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
276 * struct cftype: handler definitions for cgroup control files
278 * When reading/writing to a file:
279 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
280 * - the 'cftype' of the file is file->f_dentry->d_fsdata
284 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
285 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
287 #define MAX_CFTYPE_NAME 64
291 * By convention, the name should begin with the name of the
292 * subsystem, followed by a period. Zero length string indicates
293 * end of cftype array.
295 char name
[MAX_CFTYPE_NAME
];
298 * If not 0, file mode is set to this value, otherwise it will
299 * be figured out automatically
304 * If non-zero, defines the maximum length of string that can
305 * be passed to write_string; defaults to 64
307 size_t max_write_len
;
312 int (*open
)(struct inode
*inode
, struct file
*file
);
313 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
315 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
317 * read_u64() is a shortcut for the common case of returning a
318 * single integer. Use it in place of read()
320 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
322 * read_s64() is a signed version of read_u64()
324 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
326 * read_map() is used for defining a map of key/value
327 * pairs. It should call cb->fill(cb, key, value) for each
328 * entry. The key/value pairs (and their ordering) should not
329 * change between reboots.
331 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
332 struct cgroup_map_cb
*cb
);
334 * read_seq_string() is used for outputting a simple sequence
337 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
340 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
342 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
345 * write_u64() is a shortcut for the common case of accepting
346 * a single integer (as parsed by simple_strtoull) from
347 * userspace. Use in place of write(); return 0 or error.
349 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
351 * write_s64() is a signed version of write_u64()
353 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
356 * write_string() is passed a nul-terminated kernelspace
357 * buffer of maximum length determined by max_write_len.
358 * Returns 0 or -ve error code.
360 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
363 * trigger() callback can be used to get some kick from the
364 * userspace, when the actual string written is not important
365 * at all. The private field can be used to determine the
366 * kick type for multiplexing.
368 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
370 int (*release
)(struct inode
*inode
, struct file
*file
);
373 * register_event() callback will be used to add new userspace
374 * waiter for changes related to the cftype. Implement it if
375 * you want to provide this functionality. Use eventfd_signal()
376 * on eventfd to send notification to userspace.
378 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
379 struct eventfd_ctx
*eventfd
, const char *args
);
381 * unregister_event() callback will be called when userspace
382 * closes the eventfd or on cgroup removing.
383 * This callback must be implemented, if you want provide
384 * notification functionality.
386 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
387 struct eventfd_ctx
*eventfd
);
391 * cftype_sets describe cftypes belonging to a subsystem and are chained at
392 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
393 * terminated by zero length name.
396 struct list_head node
; /* chained at subsys->cftsets */
397 const struct cftype
*cfts
;
400 struct cgroup_scanner
{
402 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
403 void (*process_task
)(struct task_struct
*p
,
404 struct cgroup_scanner
*scan
);
405 struct ptr_heap
*heap
;
409 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
);
410 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
);
412 int cgroup_is_removed(const struct cgroup
*cgrp
);
414 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
416 int cgroup_task_count(const struct cgroup
*cgrp
);
418 /* Return true if cgrp is a descendant of the task's cgroup */
419 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
422 * When the subsys has to access css and may add permanent refcnt to css,
423 * it should take care of racy conditions with rmdir(). Following set of
424 * functions, is for stop/restart rmdir if necessary.
425 * Because these will call css_get/put, "css" should be alive css.
427 * cgroup_exclude_rmdir();
428 * ...do some jobs which may access arbitrary empty cgroup
429 * cgroup_release_and_wakeup_rmdir();
431 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
432 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
435 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
);
436 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
);
439 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
442 struct cgroup_taskset
;
443 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
444 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
445 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
446 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
449 * cgroup_taskset_for_each - iterate cgroup_taskset
450 * @task: the loop cursor
451 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
452 * @tset: taskset to iterate
454 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
455 for ((task) = cgroup_taskset_first((tset)); (task); \
456 (task) = cgroup_taskset_next((tset))) \
457 if (!(skip_cgrp) || \
458 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
461 * Control Group subsystem type.
462 * See Documentation/cgroups/cgroups.txt for details
465 struct cgroup_subsys
{
466 struct cgroup_subsys_state
*(*create
)(struct cgroup
*cgrp
);
467 int (*pre_destroy
)(struct cgroup
*cgrp
);
468 void (*destroy
)(struct cgroup
*cgrp
);
469 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
470 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
471 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
472 void (*fork
)(struct task_struct
*task
);
473 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
474 struct task_struct
*task
);
475 void (*post_clone
)(struct cgroup
*cgrp
);
476 void (*bind
)(struct cgroup
*root
);
483 * True if this subsys uses ID. ID is not available before cgroup_init()
484 * (not available in early_init time.)
489 * If %true, cgroup removal will try to clear css refs by retrying
490 * ss->pre_destroy() until there's no css ref left. This behavior
491 * is strictly for backward compatibility and will be removed as
492 * soon as the current user (memcg) is updated.
494 * If %false, ss->pre_destroy() can't fail and cgroup removal won't
495 * wait for css refs to drop to zero before proceeding.
497 bool __DEPRECATED_clear_css_refs
;
499 #define MAX_CGROUP_TYPE_NAMELEN 32
503 * Protects sibling/children links of cgroups in this
504 * hierarchy, plus protects which hierarchy (or none) the
505 * subsystem is a part of (i.e. root/sibling). To avoid
506 * potential deadlocks, the following operations should not be
507 * undertaken while holding any hierarchy_mutex:
509 * - allocating memory
510 * - initiating hotplug events
512 struct mutex hierarchy_mutex
;
513 struct lock_class_key subsys_key
;
516 * Link to parent, and list entry in parent's children.
517 * Protected by this->hierarchy_mutex and cgroup_lock()
519 struct cgroupfs_root
*root
;
520 struct list_head sibling
;
521 /* used when use_id == true */
525 /* list of cftype_sets */
526 struct list_head cftsets
;
528 /* base cftypes, automatically [de]registered with subsys itself */
529 struct cftype
*base_cftypes
;
530 struct cftype_set base_cftset
;
532 /* should be defined only by modular subsystems */
533 struct module
*module
;
536 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
537 #include <linux/cgroup_subsys.h>
540 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
541 struct cgroup
*cgrp
, int subsys_id
)
543 return cgrp
->subsys
[subsys_id
];
547 * function to get the cgroup_subsys_state which allows for extra
548 * rcu_dereference_check() conditions, such as locks used during the
549 * cgroup_subsys::attach() methods.
551 #define task_subsys_state_check(task, subsys_id, __c) \
552 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
553 lockdep_is_held(&task->alloc_lock) || \
554 cgroup_lock_is_held() || (__c))
556 static inline struct cgroup_subsys_state
*
557 task_subsys_state(struct task_struct
*task
, int subsys_id
)
559 return task_subsys_state_check(task
, subsys_id
, false);
562 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
565 return task_subsys_state(task
, subsys_id
)->cgroup
;
568 /* A cgroup_iter should be treated as an opaque object */
570 struct list_head
*cg_link
;
571 struct list_head
*task
;
575 * To iterate across the tasks in a cgroup:
577 * 1) call cgroup_iter_start to initialize an iterator
579 * 2) call cgroup_iter_next() to retrieve member tasks until it
580 * returns NULL or until you want to end the iteration
582 * 3) call cgroup_iter_end() to destroy the iterator.
584 * Or, call cgroup_scan_tasks() to iterate through every task in a
585 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
586 * the test_task() callback, but not while calling the process_task()
589 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
590 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
591 struct cgroup_iter
*it
);
592 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
593 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
594 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
595 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
598 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
599 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
600 * CSS ID is assigned at cgroup allocation (create) automatically
601 * and removed when subsys calls free_css_id() function. This is because
602 * the lifetime of cgroup_subsys_state is subsys's matter.
604 * Looking up and scanning function should be called under rcu_read_lock().
605 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
606 * But the css returned by this routine can be "not populated yet" or "being
607 * destroyed". The caller should check css and cgroup's status.
611 * Typically Called at ->destroy(), or somewhere the subsys frees
612 * cgroup_subsys_state.
614 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
616 /* Find a cgroup_subsys_state which has given ID */
618 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
621 * Get a cgroup whose id is greater than or equal to id under tree of root.
622 * Returning a cgroup_subsys_state or NULL.
624 struct cgroup_subsys_state
*css_get_next(struct cgroup_subsys
*ss
, int id
,
625 struct cgroup_subsys_state
*root
, int *foundid
);
627 /* Returns true if root is ancestor of cg */
628 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
629 const struct cgroup_subsys_state
*root
);
631 /* Get id and depth of css */
632 unsigned short css_id(struct cgroup_subsys_state
*css
);
633 unsigned short css_depth(struct cgroup_subsys_state
*css
);
634 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
636 #else /* !CONFIG_CGROUPS */
638 static inline int cgroup_init_early(void) { return 0; }
639 static inline int cgroup_init(void) { return 0; }
640 static inline void cgroup_fork(struct task_struct
*p
) {}
641 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
642 static inline void cgroup_post_fork(struct task_struct
*p
) {}
643 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
645 static inline void cgroup_lock(void) {}
646 static inline void cgroup_unlock(void) {}
647 static inline int cgroupstats_build(struct cgroupstats
*stats
,
648 struct dentry
*dentry
)
653 /* No cgroups - nothing to do */
654 static inline int cgroup_attach_task_all(struct task_struct
*from
,
655 struct task_struct
*t
)
660 #endif /* !CONFIG_CGROUPS */
662 #endif /* _LINUX_CGROUP_H */