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>
20 #include <linux/xattr.h>
30 extern int cgroup_init_early(void);
31 extern int cgroup_init(void);
32 extern void cgroup_lock(void);
33 extern int cgroup_lock_is_held(void);
34 extern bool cgroup_lock_live_group(struct cgroup
*cgrp
);
35 extern void cgroup_unlock(void);
36 extern void cgroup_fork(struct task_struct
*p
);
37 extern void cgroup_fork_callbacks(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
, /* This CSS is the root of the subsystem */
85 CSS_REMOVED
, /* This CSS is dead */
86 CSS_CLEAR_CSS_REFS
, /* @ss->__DEPRECATED_clear_css_refs */
89 /* Caller must verify that the css is not for root cgroup */
90 static inline void __css_get(struct cgroup_subsys_state
*css
, int count
)
92 atomic_add(count
, &css
->refcnt
);
96 * Call css_get() to hold a reference on the css; it can be used
97 * for a reference obtained via:
98 * - an existing ref-counted reference to the css
99 * - task->cgroups for a locked task
102 static inline void css_get(struct cgroup_subsys_state
*css
)
104 /* We don't need to reference count the root state */
105 if (!test_bit(CSS_ROOT
, &css
->flags
))
109 static inline bool css_is_removed(struct cgroup_subsys_state
*css
)
111 return test_bit(CSS_REMOVED
, &css
->flags
);
115 * Call css_tryget() to take a reference on a css if your existing
116 * (known-valid) reference isn't already ref-counted. Returns false if
117 * the css has been destroyed.
120 extern bool __css_tryget(struct cgroup_subsys_state
*css
);
121 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
123 if (test_bit(CSS_ROOT
, &css
->flags
))
125 return __css_tryget(css
);
129 * css_put() should be called to release a reference taken by
130 * css_get() or css_tryget()
133 extern void __css_put(struct cgroup_subsys_state
*css
);
134 static inline void css_put(struct cgroup_subsys_state
*css
)
136 if (!test_bit(CSS_ROOT
, &css
->flags
))
140 /* bits in struct cgroup flags field */
142 /* Control Group is dead */
145 * Control Group has previously had a child cgroup or a task,
146 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
149 /* Control Group requires release notifications to userspace */
150 CGRP_NOTIFY_ON_RELEASE
,
152 * A thread in rmdir() is wating for this cgroup.
156 * Clone cgroup values when creating a new child cgroup
162 unsigned long flags
; /* "unsigned long" so bitops work */
165 * count users of this cgroup. >0 means busy, but doesn't
166 * necessarily indicate the number of tasks in the cgroup
171 * We link our 'sibling' struct into our parent's 'children'.
172 * Our children link their 'sibling' into our 'children'.
174 struct list_head sibling
; /* my parent's children */
175 struct list_head children
; /* my children */
176 struct list_head files
; /* my files */
178 struct cgroup
*parent
; /* my parent */
179 struct dentry __rcu
*dentry
; /* cgroup fs entry, RCU protected */
181 /* Private pointers for each registered subsystem */
182 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
184 struct cgroupfs_root
*root
;
185 struct cgroup
*top_cgroup
;
188 * List of cg_cgroup_links pointing at css_sets with
189 * tasks in this cgroup. Protected by css_set_lock
191 struct list_head css_sets
;
193 struct list_head allcg_node
; /* cgroupfs_root->allcg_list */
194 struct list_head cft_q_node
; /* used during cftype add/rm */
197 * Linked list running through all cgroups that can
198 * potentially be reaped by the release agent. Protected by
201 struct list_head release_list
;
204 * list of pidlists, up to two for each namespace (one for procs, one
205 * for tasks); created on demand.
207 struct list_head pidlists
;
208 struct mutex pidlist_mutex
;
210 /* For RCU-protected deletion */
211 struct rcu_head rcu_head
;
213 /* List of events which userspace want to receive */
214 struct list_head event_list
;
215 spinlock_t event_list_lock
;
217 /* directory xattrs */
218 struct simple_xattrs xattrs
;
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
;
313 struct simple_xattrs xattrs
;
315 int (*open
)(struct inode
*inode
, struct file
*file
);
316 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
318 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
320 * read_u64() is a shortcut for the common case of returning a
321 * single integer. Use it in place of read()
323 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
325 * read_s64() is a signed version of read_u64()
327 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
329 * read_map() is used for defining a map of key/value
330 * pairs. It should call cb->fill(cb, key, value) for each
331 * entry. The key/value pairs (and their ordering) should not
332 * change between reboots.
334 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
335 struct cgroup_map_cb
*cb
);
337 * read_seq_string() is used for outputting a simple sequence
340 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
343 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
345 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
348 * write_u64() is a shortcut for the common case of accepting
349 * a single integer (as parsed by simple_strtoull) from
350 * userspace. Use in place of write(); return 0 or error.
352 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
354 * write_s64() is a signed version of write_u64()
356 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
359 * write_string() is passed a nul-terminated kernelspace
360 * buffer of maximum length determined by max_write_len.
361 * Returns 0 or -ve error code.
363 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
366 * trigger() callback can be used to get some kick from the
367 * userspace, when the actual string written is not important
368 * at all. The private field can be used to determine the
369 * kick type for multiplexing.
371 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
373 int (*release
)(struct inode
*inode
, struct file
*file
);
376 * register_event() callback will be used to add new userspace
377 * waiter for changes related to the cftype. Implement it if
378 * you want to provide this functionality. Use eventfd_signal()
379 * on eventfd to send notification to userspace.
381 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
382 struct eventfd_ctx
*eventfd
, const char *args
);
384 * unregister_event() callback will be called when userspace
385 * closes the eventfd or on cgroup removing.
386 * This callback must be implemented, if you want provide
387 * notification functionality.
389 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
390 struct eventfd_ctx
*eventfd
);
394 * cftype_sets describe cftypes belonging to a subsystem and are chained at
395 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
396 * terminated by zero length name.
399 struct list_head node
; /* chained at subsys->cftsets */
403 struct cgroup_scanner
{
405 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
406 void (*process_task
)(struct task_struct
*p
,
407 struct cgroup_scanner
*scan
);
408 struct ptr_heap
*heap
;
412 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
413 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
415 int cgroup_is_removed(const struct cgroup
*cgrp
);
417 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
419 int cgroup_task_count(const struct cgroup
*cgrp
);
421 /* Return true if cgrp is a descendant of the task's cgroup */
422 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
);
425 * When the subsys has to access css and may add permanent refcnt to css,
426 * it should take care of racy conditions with rmdir(). Following set of
427 * functions, is for stop/restart rmdir if necessary.
428 * Because these will call css_get/put, "css" should be alive css.
430 * cgroup_exclude_rmdir();
431 * ...do some jobs which may access arbitrary empty cgroup
432 * cgroup_release_and_wakeup_rmdir();
434 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
435 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
438 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
);
439 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
);
442 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
445 struct cgroup_taskset
;
446 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
447 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
448 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
449 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
452 * cgroup_taskset_for_each - iterate cgroup_taskset
453 * @task: the loop cursor
454 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
455 * @tset: taskset to iterate
457 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
458 for ((task) = cgroup_taskset_first((tset)); (task); \
459 (task) = cgroup_taskset_next((tset))) \
460 if (!(skip_cgrp) || \
461 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
464 * Control Group subsystem type.
465 * See Documentation/cgroups/cgroups.txt for details
468 struct cgroup_subsys
{
469 struct cgroup_subsys_state
*(*create
)(struct cgroup
*cgrp
);
470 int (*pre_destroy
)(struct cgroup
*cgrp
);
471 void (*destroy
)(struct cgroup
*cgrp
);
472 int (*can_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
473 void (*cancel_attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
474 void (*attach
)(struct cgroup
*cgrp
, struct cgroup_taskset
*tset
);
475 void (*fork
)(struct task_struct
*task
);
476 void (*exit
)(struct cgroup
*cgrp
, struct cgroup
*old_cgrp
,
477 struct task_struct
*task
);
478 void (*post_clone
)(struct cgroup
*cgrp
);
479 void (*bind
)(struct cgroup
*root
);
486 * True if this subsys uses ID. ID is not available before cgroup_init()
487 * (not available in early_init time.)
492 * If %true, cgroup removal will try to clear css refs by retrying
493 * ss->pre_destroy() until there's no css ref left. This behavior
494 * is strictly for backward compatibility and will be removed as
495 * soon as the current user (memcg) is updated.
497 * If %false, ss->pre_destroy() can't fail and cgroup removal won't
498 * wait for css refs to drop to zero before proceeding.
500 bool __DEPRECATED_clear_css_refs
;
503 * If %false, this subsystem is properly hierarchical -
504 * configuration, resource accounting and restriction on a parent
505 * cgroup cover those of its children. If %true, hierarchy support
506 * is broken in some ways - some subsystems ignore hierarchy
507 * completely while others are only implemented half-way.
509 * It's now disallowed to create nested cgroups if the subsystem is
510 * broken and cgroup core will emit a warning message on such
511 * cases. Eventually, all subsystems will be made properly
512 * hierarchical and this will go away.
514 bool broken_hierarchy
;
515 bool warned_broken_hierarchy
;
517 #define MAX_CGROUP_TYPE_NAMELEN 32
521 * Link to parent, and list entry in parent's children.
522 * Protected by cgroup_lock()
524 struct cgroupfs_root
*root
;
525 struct list_head sibling
;
526 /* used when use_id == true */
530 /* list of cftype_sets */
531 struct list_head cftsets
;
533 /* base cftypes, automatically [de]registered with subsys itself */
534 struct cftype
*base_cftypes
;
535 struct cftype_set base_cftset
;
537 /* should be defined only by modular subsystems */
538 struct module
*module
;
541 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
542 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
543 #include <linux/cgroup_subsys.h>
544 #undef IS_SUBSYS_ENABLED
547 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
548 struct cgroup
*cgrp
, int subsys_id
)
550 return cgrp
->subsys
[subsys_id
];
554 * function to get the cgroup_subsys_state which allows for extra
555 * rcu_dereference_check() conditions, such as locks used during the
556 * cgroup_subsys::attach() methods.
558 #define task_subsys_state_check(task, subsys_id, __c) \
559 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
560 lockdep_is_held(&task->alloc_lock) || \
561 cgroup_lock_is_held() || (__c))
563 static inline struct cgroup_subsys_state
*
564 task_subsys_state(struct task_struct
*task
, int subsys_id
)
566 return task_subsys_state_check(task
, subsys_id
, false);
569 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
572 return task_subsys_state(task
, subsys_id
)->cgroup
;
575 /* A cgroup_iter should be treated as an opaque object */
577 struct list_head
*cg_link
;
578 struct list_head
*task
;
582 * To iterate across the tasks in a cgroup:
584 * 1) call cgroup_iter_start to initialize an iterator
586 * 2) call cgroup_iter_next() to retrieve member tasks until it
587 * returns NULL or until you want to end the iteration
589 * 3) call cgroup_iter_end() to destroy the iterator.
591 * Or, call cgroup_scan_tasks() to iterate through every task in a
592 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
593 * the test_task() callback, but not while calling the process_task()
596 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
597 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
598 struct cgroup_iter
*it
);
599 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
600 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
601 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
602 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
605 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
606 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
607 * CSS ID is assigned at cgroup allocation (create) automatically
608 * and removed when subsys calls free_css_id() function. This is because
609 * the lifetime of cgroup_subsys_state is subsys's matter.
611 * Looking up and scanning function should be called under rcu_read_lock().
612 * Taking cgroup_mutex is not necessary for following calls.
613 * But the css returned by this routine can be "not populated yet" or "being
614 * destroyed". The caller should check css and cgroup's status.
618 * Typically Called at ->destroy(), or somewhere the subsys frees
619 * cgroup_subsys_state.
621 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
623 /* Find a cgroup_subsys_state which has given ID */
625 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
628 * Get a cgroup whose id is greater than or equal to id under tree of root.
629 * Returning a cgroup_subsys_state or NULL.
631 struct cgroup_subsys_state
*css_get_next(struct cgroup_subsys
*ss
, int id
,
632 struct cgroup_subsys_state
*root
, int *foundid
);
634 /* Returns true if root is ancestor of cg */
635 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
636 const struct cgroup_subsys_state
*root
);
638 /* Get id and depth of css */
639 unsigned short css_id(struct cgroup_subsys_state
*css
);
640 unsigned short css_depth(struct cgroup_subsys_state
*css
);
641 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
643 #else /* !CONFIG_CGROUPS */
645 static inline int cgroup_init_early(void) { return 0; }
646 static inline int cgroup_init(void) { return 0; }
647 static inline void cgroup_fork(struct task_struct
*p
) {}
648 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
649 static inline void cgroup_post_fork(struct task_struct
*p
) {}
650 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
652 static inline void cgroup_lock(void) {}
653 static inline void cgroup_unlock(void) {}
654 static inline int cgroupstats_build(struct cgroupstats
*stats
,
655 struct dentry
*dentry
)
660 /* No cgroups - nothing to do */
661 static inline int cgroup_attach_task_all(struct task_struct
*from
,
662 struct task_struct
*t
)
667 #endif /* !CONFIG_CGROUPS */
669 #endif /* _LINUX_CGROUP_H */