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>
28 extern int cgroup_init_early(void);
29 extern int cgroup_init(void);
30 extern void cgroup_lock(void);
31 extern int cgroup_lock_is_held(void);
32 extern bool cgroup_lock_live_group(struct cgroup
*cgrp
);
33 extern void cgroup_unlock(void);
34 extern void cgroup_fork(struct task_struct
*p
);
35 extern void cgroup_fork_callbacks(struct task_struct
*p
);
36 extern void cgroup_post_fork(struct task_struct
*p
);
37 extern void cgroup_exit(struct task_struct
*p
, int run_callbacks
);
38 extern int cgroupstats_build(struct cgroupstats
*stats
,
39 struct dentry
*dentry
);
40 extern int cgroup_load_subsys(struct cgroup_subsys
*ss
);
41 extern void cgroup_unload_subsys(struct cgroup_subsys
*ss
);
43 extern const struct file_operations proc_cgroup_operations
;
45 /* Define the enumeration of all builtin cgroup subsystems */
46 #define SUBSYS(_x) _x ## _subsys_id,
47 enum cgroup_subsys_id
{
48 #include <linux/cgroup_subsys.h>
49 CGROUP_BUILTIN_SUBSYS_COUNT
53 * This define indicates the maximum number of subsystems that can be loaded
54 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
55 * track of all of them.
57 #define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
59 /* Per-subsystem/per-cgroup state maintained by the system. */
60 struct cgroup_subsys_state
{
62 * The cgroup that this subsystem is attached to. Useful
63 * for subsystems that want to know about the cgroup
66 struct cgroup
*cgroup
;
69 * State maintained by the cgroup system to allow subsystems
70 * to be "busy". Should be accessed via css_get(),
71 * css_tryget() and and css_put().
77 /* ID for this css, if possible */
78 struct css_id __rcu
*id
;
81 /* bits in struct cgroup_subsys_state flags field */
83 CSS_ROOT
, /* This CSS is the root of the subsystem */
84 CSS_REMOVED
, /* This CSS is dead */
87 /* Caller must verify that the css is not for root cgroup */
88 static inline void __css_get(struct cgroup_subsys_state
*css
, int count
)
90 atomic_add(count
, &css
->refcnt
);
94 * Call css_get() to hold a reference on the css; it can be used
95 * for a reference obtained via:
96 * - an existing ref-counted reference to the css
97 * - task->cgroups for a locked task
100 static inline void css_get(struct cgroup_subsys_state
*css
)
102 /* We don't need to reference count the root state */
103 if (!test_bit(CSS_ROOT
, &css
->flags
))
107 static inline bool css_is_removed(struct cgroup_subsys_state
*css
)
109 return test_bit(CSS_REMOVED
, &css
->flags
);
113 * Call css_tryget() to take a reference on a css if your existing
114 * (known-valid) reference isn't already ref-counted. Returns false if
115 * the css has been destroyed.
118 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
120 if (test_bit(CSS_ROOT
, &css
->flags
))
122 while (!atomic_inc_not_zero(&css
->refcnt
)) {
123 if (test_bit(CSS_REMOVED
, &css
->flags
))
131 * css_put() should be called to release a reference taken by
132 * css_get() or css_tryget()
135 extern void __css_put(struct cgroup_subsys_state
*css
, int count
);
136 static inline void css_put(struct cgroup_subsys_state
*css
)
138 if (!test_bit(CSS_ROOT
, &css
->flags
))
142 /* bits in struct cgroup flags field */
144 /* Control Group is dead */
147 * Control Group has previously had a child cgroup or a task,
148 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
151 /* Control Group requires release notifications to userspace */
152 CGRP_NOTIFY_ON_RELEASE
,
154 * A thread in rmdir() is wating for this cgroup.
158 * Clone cgroup values when creating a new child cgroup
164 unsigned long flags
; /* "unsigned long" so bitops work */
167 * count users of this cgroup. >0 means busy, but doesn't
168 * necessarily indicate the number of tasks in the cgroup
173 * We link our 'sibling' struct into our parent's 'children'.
174 * Our children link their 'sibling' into our 'children'.
176 struct list_head sibling
; /* my parent's children */
177 struct list_head children
; /* my children */
179 struct cgroup
*parent
; /* my parent */
180 struct dentry __rcu
*dentry
; /* cgroup fs entry, RCU protected */
182 /* Private pointers for each registered subsystem */
183 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
185 struct cgroupfs_root
*root
;
186 struct cgroup
*top_cgroup
;
189 * List of cg_cgroup_links pointing at css_sets with
190 * tasks in this cgroup. Protected by css_set_lock
192 struct list_head css_sets
;
195 * Linked list running through all cgroups that can
196 * potentially be reaped by the release agent. Protected by
199 struct list_head release_list
;
202 * list of pidlists, up to two for each namespace (one for procs, one
203 * for tasks); created on demand.
205 struct list_head pidlists
;
206 struct mutex pidlist_mutex
;
208 /* For RCU-protected deletion */
209 struct rcu_head rcu_head
;
211 /* List of events which userspace want to receive */
212 struct list_head event_list
;
213 spinlock_t event_list_lock
;
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
278 #define MAX_CFTYPE_NAME 64
281 * By convention, the name should begin with the name of the
282 * subsystem, followed by a period
284 char name
[MAX_CFTYPE_NAME
];
287 * If not 0, file mode is set to this value, otherwise it will
288 * be figured out automatically
293 * If non-zero, defines the maximum length of string that can
294 * be passed to write_string; defaults to 64
296 size_t max_write_len
;
298 int (*open
)(struct inode
*inode
, struct file
*file
);
299 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
301 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
303 * read_u64() is a shortcut for the common case of returning a
304 * single integer. Use it in place of read()
306 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
308 * read_s64() is a signed version of read_u64()
310 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
312 * read_map() is used for defining a map of key/value
313 * pairs. It should call cb->fill(cb, key, value) for each
314 * entry. The key/value pairs (and their ordering) should not
315 * change between reboots.
317 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
318 struct cgroup_map_cb
*cb
);
320 * read_seq_string() is used for outputting a simple sequence
323 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
326 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
328 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
331 * write_u64() is a shortcut for the common case of accepting
332 * a single integer (as parsed by simple_strtoull) from
333 * userspace. Use in place of write(); return 0 or error.
335 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
337 * write_s64() is a signed version of write_u64()
339 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
342 * write_string() is passed a nul-terminated kernelspace
343 * buffer of maximum length determined by max_write_len.
344 * Returns 0 or -ve error code.
346 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
349 * trigger() callback can be used to get some kick from the
350 * userspace, when the actual string written is not important
351 * at all. The private field can be used to determine the
352 * kick type for multiplexing.
354 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
356 int (*release
)(struct inode
*inode
, struct file
*file
);
359 * register_event() callback will be used to add new userspace
360 * waiter for changes related to the cftype. Implement it if
361 * you want to provide this functionality. Use eventfd_signal()
362 * on eventfd to send notification to userspace.
364 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
365 struct eventfd_ctx
*eventfd
, const char *args
);
367 * unregister_event() callback will be called when userspace
368 * closes the eventfd or on cgroup removing.
369 * This callback must be implemented, if you want provide
370 * notification functionality.
372 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
373 struct eventfd_ctx
*eventfd
);
376 struct cgroup_scanner
{
378 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
379 void (*process_task
)(struct task_struct
*p
,
380 struct cgroup_scanner
*scan
);
381 struct ptr_heap
*heap
;
386 * Add a new file to the given cgroup directory. Should only be
387 * called by subsystems from within a populate() method
389 int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
390 const struct cftype
*cft
);
393 * Add a set of new files to the given cgroup directory. Should
394 * only be called by subsystems from within a populate() method
396 int cgroup_add_files(struct cgroup
*cgrp
,
397 struct cgroup_subsys
*subsys
,
398 const struct cftype cft
[],
401 int cgroup_is_removed(const struct cgroup
*cgrp
);
403 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
405 int cgroup_task_count(const struct cgroup
*cgrp
);
407 /* Return true if cgrp is a descendant of the task's cgroup */
408 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
411 * When the subsys has to access css and may add permanent refcnt to css,
412 * it should take care of racy conditions with rmdir(). Following set of
413 * functions, is for stop/restart rmdir if necessary.
414 * Because these will call css_get/put, "css" should be alive css.
416 * cgroup_exclude_rmdir();
417 * ...do some jobs which may access arbitrary empty cgroup
418 * cgroup_release_and_wakeup_rmdir();
420 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
421 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
424 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
);
425 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
);
428 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
431 struct cgroup_taskset
;
432 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
433 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
434 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
435 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
438 * cgroup_taskset_for_each - iterate cgroup_taskset
439 * @task: the loop cursor
440 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
441 * @tset: taskset to iterate
443 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
444 for ((task) = cgroup_taskset_first((tset)); (task); \
445 (task) = cgroup_taskset_next((tset))) \
446 if (!(skip_cgrp) || \
447 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
450 * Control Group subsystem type.
451 * See Documentation/cgroups/cgroups.txt for details
454 struct cgroup_subsys
{
455 struct cgroup_subsys_state
*(*create
)(struct cgroup
*cgrp
);
456 int (*pre_destroy
)(struct cgroup
*cgrp
);
457 void (*destroy
)(struct cgroup
*cgrp
);
458 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
459 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
460 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
461 void (*fork
)(struct task_struct
*task
);
462 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
463 struct task_struct
*task
);
464 int (*populate
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
465 void (*post_clone
)(struct cgroup
*cgrp
);
466 void (*bind
)(struct cgroup
*root
);
473 * True if this subsys uses ID. ID is not available before cgroup_init()
474 * (not available in early_init time.)
477 #define MAX_CGROUP_TYPE_NAMELEN 32
481 * Protects sibling/children links of cgroups in this
482 * hierarchy, plus protects which hierarchy (or none) the
483 * subsystem is a part of (i.e. root/sibling). To avoid
484 * potential deadlocks, the following operations should not be
485 * undertaken while holding any hierarchy_mutex:
487 * - allocating memory
488 * - initiating hotplug events
490 struct mutex hierarchy_mutex
;
491 struct lock_class_key subsys_key
;
494 * Link to parent, and list entry in parent's children.
495 * Protected by this->hierarchy_mutex and cgroup_lock()
497 struct cgroupfs_root
*root
;
498 struct list_head sibling
;
499 /* used when use_id == true */
503 /* should be defined only by modular subsystems */
504 struct module
*module
;
507 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
508 #include <linux/cgroup_subsys.h>
511 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
512 struct cgroup
*cgrp
, int subsys_id
)
514 return cgrp
->subsys
[subsys_id
];
518 * function to get the cgroup_subsys_state which allows for extra
519 * rcu_dereference_check() conditions, such as locks used during the
520 * cgroup_subsys::attach() methods.
522 #define task_subsys_state_check(task, subsys_id, __c) \
523 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
524 lockdep_is_held(&task->alloc_lock) || \
525 cgroup_lock_is_held() || (__c))
527 static inline struct cgroup_subsys_state
*
528 task_subsys_state(struct task_struct
*task
, int subsys_id
)
530 return task_subsys_state_check(task
, subsys_id
, false);
533 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
536 return task_subsys_state(task
, subsys_id
)->cgroup
;
539 /* A cgroup_iter should be treated as an opaque object */
541 struct list_head
*cg_link
;
542 struct list_head
*task
;
546 * To iterate across the tasks in a cgroup:
548 * 1) call cgroup_iter_start to initialize an iterator
550 * 2) call cgroup_iter_next() to retrieve member tasks until it
551 * returns NULL or until you want to end the iteration
553 * 3) call cgroup_iter_end() to destroy the iterator.
555 * Or, call cgroup_scan_tasks() to iterate through every task in a
556 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
557 * the test_task() callback, but not while calling the process_task()
560 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
561 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
562 struct cgroup_iter
*it
);
563 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
564 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
565 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
566 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
569 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
570 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
571 * CSS ID is assigned at cgroup allocation (create) automatically
572 * and removed when subsys calls free_css_id() function. This is because
573 * the lifetime of cgroup_subsys_state is subsys's matter.
575 * Looking up and scanning function should be called under rcu_read_lock().
576 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
577 * But the css returned by this routine can be "not populated yet" or "being
578 * destroyed". The caller should check css and cgroup's status.
582 * Typically Called at ->destroy(), or somewhere the subsys frees
583 * cgroup_subsys_state.
585 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
587 /* Find a cgroup_subsys_state which has given ID */
589 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
592 * Get a cgroup whose id is greater than or equal to id under tree of root.
593 * Returning a cgroup_subsys_state or NULL.
595 struct cgroup_subsys_state
*css_get_next(struct cgroup_subsys
*ss
, int id
,
596 struct cgroup_subsys_state
*root
, int *foundid
);
598 /* Returns true if root is ancestor of cg */
599 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
600 const struct cgroup_subsys_state
*root
);
602 /* Get id and depth of css */
603 unsigned short css_id(struct cgroup_subsys_state
*css
);
604 unsigned short css_depth(struct cgroup_subsys_state
*css
);
605 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
607 #else /* !CONFIG_CGROUPS */
609 static inline int cgroup_init_early(void) { return 0; }
610 static inline int cgroup_init(void) { return 0; }
611 static inline void cgroup_fork(struct task_struct
*p
) {}
612 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
613 static inline void cgroup_post_fork(struct task_struct
*p
) {}
614 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
616 static inline void cgroup_lock(void) {}
617 static inline void cgroup_unlock(void) {}
618 static inline int cgroupstats_build(struct cgroupstats
*stats
,
619 struct dentry
*dentry
)
624 /* No cgroups - nothing to do */
625 static inline int cgroup_attach_task_all(struct task_struct
*from
,
626 struct task_struct
*t
)
631 #endif /* !CONFIG_CGROUPS */
633 #endif /* _LINUX_CGROUP_H */