/include/linux/cgroup.h

  1#ifndef _LINUX_CGROUP_H
  2#define _LINUX_CGROUP_H
  3/*
  4 *  cgroup interface
  5 *
  6 *  Copyright (C) 2003 BULL SA
  7 *  Copyright (C) 2004-2006 Silicon Graphics, Inc.
  8 *
  9 */
 10
 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
 20#ifdef CONFIG_CGROUPS
 21
 22struct cgroupfs_root;
 23struct cgroup_subsys;
 24struct inode;
 25struct cgroup;
 26struct css_id;
 27
 28extern int cgroup_init_early(void);
 29extern int cgroup_init(void);
 30extern void cgroup_lock(void);
 31extern int cgroup_lock_is_held(void);
 32extern bool cgroup_lock_live_group(struct cgroup *cgrp);
 33extern void cgroup_unlock(void);
 34extern void cgroup_fork(struct task_struct *p);
 35extern void cgroup_fork_callbacks(struct task_struct *p);
 36extern void cgroup_post_fork(struct task_struct *p);
 37extern void cgroup_exit(struct task_struct *p, int run_callbacks);
 38extern int cgroupstats_build(struct cgroupstats *stats,
 39				struct dentry *dentry);
 40extern int cgroup_load_subsys(struct cgroup_subsys *ss);
 41extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
 42
 43extern const struct file_operations proc_cgroup_operations;
 44
 45/* Define the enumeration of all builtin cgroup subsystems */
 46#define SUBSYS(_x) _x ## _subsys_id,
 47enum cgroup_subsys_id {
 48#include <linux/cgroup_subsys.h>
 49	CGROUP_BUILTIN_SUBSYS_COUNT
 50};
 51#undef SUBSYS
 52/*
 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.
 56 */
 57#define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
 58
 59/* Per-subsystem/per-cgroup state maintained by the system. */
 60struct cgroup_subsys_state {
 61	/*
 62	 * The cgroup that this subsystem is attached to. Useful
 63	 * for subsystems that want to know about the cgroup
 64	 * hierarchy structure
 65	 */
 66	struct cgroup *cgroup;
 67
 68	/*
 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().
 72	 */
 73
 74	atomic_t refcnt;
 75
 76	unsigned long flags;
 77	/* ID for this css, if possible */
 78	struct css_id __rcu *id;
 79};
 80
 81/* bits in struct cgroup_subsys_state flags field */
 82enum {
 83	CSS_ROOT, /* This CSS is the root of the subsystem */
 84	CSS_REMOVED, /* This CSS is dead */
 85};
 86
 87/*
 88 * Call css_get() to hold a reference on the css; it can be used
 89 * for a reference obtained via:
 90 * - an existing ref-counted reference to the css
 91 * - task->cgroups for a locked task
 92 */
 93
 94extern void __css_get(struct cgroup_subsys_state *css, int count);
 95static inline void css_get(struct cgroup_subsys_state *css)
 96{
 97	/* We don't need to reference count the root state */
 98	if (!test_bit(CSS_ROOT, &css->flags))
 99		__css_get(css, 1);
100}
101
102static inline bool css_is_removed(struct cgroup_subsys_state *css)
103{
104	return test_bit(CSS_REMOVED, &css->flags);
105}
106
107/*
108 * Call css_tryget() to take a reference on a css if your existing
109 * (known-valid) reference isn't already ref-counted. Returns false if
110 * the css has been destroyed.
111 */
112
113static inline bool css_tryget(struct cgroup_subsys_state *css)
114{
115	if (test_bit(CSS_ROOT, &css->flags))
116		return true;
117	while (!atomic_inc_not_zero(&css->refcnt)) {
118		if (test_bit(CSS_REMOVED, &css->flags))
119			return false;
120		cpu_relax();
121	}
122	return true;
123}
124
125/*
126 * css_put() should be called to release a reference taken by
127 * css_get() or css_tryget()
128 */
129
130extern void __css_put(struct cgroup_subsys_state *css, int count);
131static inline void css_put(struct cgroup_subsys_state *css)
132{
133	if (!test_bit(CSS_ROOT, &css->flags))
134		__css_put(css, 1);
135}
136
137/* bits in struct cgroup flags field */
138enum {
139	/* Control Group is dead */
140	CGRP_REMOVED,
141	/* Control Group has ever had a child cgroup or a task */
142	CGRP_RELEASABLE,
143	/* Control Group requires release notifications to userspace */
144	CGRP_NOTIFY_ON_RELEASE,
145	/*
146	 * A thread in rmdir() is wating for this cgroup.
147	 */
148	CGRP_WAIT_ON_RMDIR,
149	/*
150	 * Clone cgroup values when creating a new child cgroup
151	 */
152	CGRP_CLONE_CHILDREN,
153};
154
155/* which pidlist file are we talking about? */
156enum cgroup_filetype {
157	CGROUP_FILE_PROCS,
158	CGROUP_FILE_TASKS,
159};
160
161/*
162 * A pidlist is a list of pids that virtually represents the contents of one
163 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
164 * a pair (one each for procs, tasks) for each pid namespace that's relevant
165 * to the cgroup.
166 */
167struct cgroup_pidlist {
168	/*
169	 * used to find which pidlist is wanted. doesn't change as long as
170	 * this particular list stays in the list.
171	 */
172	struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
173	/* array of xids */
174	pid_t *list;
175	/* how many elements the above list has */
176	int length;
177	/* how many files are using the current array */
178	int use_count;
179	/* each of these stored in a list by its cgroup */
180	struct list_head links;
181	/* pointer to the cgroup we belong to, for list removal purposes */
182	struct cgroup *owner;
183	/* protects the other fields */
184	struct rw_semaphore mutex;
185};
186
187struct cgroup {
188	unsigned long flags;		/* "unsigned long" so bitops work */
189
190	/*
191	 * count users of this cgroup. >0 means busy, but doesn't
192	 * necessarily indicate the number of tasks in the cgroup
193	 */
194	atomic_t count;
195
196	/*
197	 * We link our 'sibling' struct into our parent's 'children'.
198	 * Our children link their 'sibling' into our 'children'.
199	 */
200	struct list_head sibling;	/* my parent's children */
201	struct list_head children;	/* my children */
202
203	struct cgroup *parent;		/* my parent */
204	struct dentry __rcu *dentry;	/* cgroup fs entry, RCU protected */
205
206	/* Private pointers for each registered subsystem */
207	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
208
209	struct cgroupfs_root *root;
210	struct cgroup *top_cgroup;
211
212	/*
213	 * List of cg_cgroup_links pointing at css_sets with
214	 * tasks in this cgroup. Protected by css_set_lock
215	 */
216	struct list_head css_sets;
217
218	/*
219	 * Linked list running through all cgroups that can
220	 * potentially be reaped by the release agent. Protected by
221	 * release_list_lock
222	 */
223	struct list_head release_list;
224
225	/*
226	 * list of pidlists, up to two for each namespace (one for procs, one
227	 * for tasks); created on demand.
228	 */
229	struct list_head pidlists;
230	struct mutex pidlist_mutex;
231
232	/* For RCU-protected deletion */
233	struct rcu_head rcu_head;
234
235	/* List of events which userspace want to receive */
236	struct list_head event_list;
237	spinlock_t event_list_lock;
238};
239
240/*
241 * A css_set is a structure holding pointers to a set of
242 * cgroup_subsys_state objects. This saves space in the task struct
243 * object and speeds up fork()/exit(), since a single inc/dec and a
244 * list_add()/del() can bump the reference count on the entire cgroup
245 * set for a task.
246 */
247
248struct css_set {
249
250	/* Reference count */
251	atomic_t refcount;
252
253	/*
254	 * List running through all cgroup groups in the same hash
255	 * slot. Protected by css_set_lock
256	 */
257	struct hlist_node hlist;
258
259	/*
260	 * List running through all tasks using this cgroup
261	 * group. Protected by css_set_lock
262	 */
263	struct list_head tasks;
264
265	/*
266	 * List of cg_cgroup_link objects on link chains from
267	 * cgroups referenced from this css_set. Protected by
268	 * css_set_lock
269	 */
270	struct list_head cg_links;
271
272	/*
273	 * Set of subsystem states, one for each subsystem. This array
274	 * is immutable after creation apart from the init_css_set
275	 * during subsystem registration (at boot time) and modular subsystem
276	 * loading/unloading.
277	 */
278	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
279
280	/* For RCU-protected deletion */
281	struct rcu_head rcu_head;
282	struct work_struct work;
283};
284
285/*
286 * cgroup_map_cb is an abstract callback API for reporting map-valued
287 * control files
288 */
289
290struct cgroup_map_cb {
291	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
292	void *state;
293};
294
295/*
296 * struct cftype: handler definitions for cgroup control files
297 *
298 * When reading/writing to a file:
299 *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
300 *	- the 'cftype' of the file is file->f_dentry->d_fsdata
301 */
302
303#define MAX_CFTYPE_NAME 64
304struct cftype {
305	/*
306	 * By convention, the name should begin with the name of the
307	 * subsystem, followed by a period
308	 */
309	char name[MAX_CFTYPE_NAME];
310	int private;
311	/*
312	 * If not 0, file mode is set to this value, otherwise it will
313	 * be figured out automatically
314	 */
315	mode_t mode;
316
317	/*
318	 * If non-zero, defines the maximum length of string that can
319	 * be passed to write_string; defaults to 64
320	 */
321	size_t max_write_len;
322
323	int (*open)(struct inode *inode, struct file *file);
324	ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
325			struct file *file,
326			char __user *buf, size_t nbytes, loff_t *ppos);
327	/*
328	 * read_u64() is a shortcut for the common case of returning a
329	 * single integer. Use it in place of read()
330	 */
331	u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
332	/*
333	 * read_s64() is a signed version of read_u64()
334	 */
335	s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
336	/*
337	 * read_map() is used for defining a map of key/value
338	 * pairs. It should call cb->fill(cb, key, value) for each
339	 * entry. The key/value pairs (and their ordering) should not
340	 * change between reboots.
341	 */
342	int (*read_map)(struct cgroup *cont, struct cftype *cft,
343			struct cgroup_map_cb *cb);
344	/*
345	 * read_seq_string() is used for outputting a simple sequence
346	 * using seqfile.
347	 */
348	int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
349			       struct seq_file *m);
350
351	ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
352			 struct file *file,
353			 const char __user *buf, size_t nbytes, loff_t *ppos);
354
355	/*
356	 * write_u64() is a shortcut for the common case of accepting
357	 * a single integer (as parsed by simple_strtoull) from
358	 * userspace. Use in place of write(); return 0 or error.
359	 */
360	int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
361	/*
362	 * write_s64() is a signed version of write_u64()
363	 */
364	int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
365
366	/*
367	 * write_string() is passed a nul-terminated kernelspace
368	 * buffer of maximum length determined by max_write_len.
369	 * Returns 0 or -ve error code.
370	 */
371	int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
372			    const char *buffer);
373	/*
374	 * trigger() callback can be used to get some kick from the
375	 * userspace, when the actual string written is not important
376	 * at all. The private field can be used to determine the
377	 * kick type for multiplexing.
378	 */
379	int (*trigger)(struct cgroup *cgrp, unsigned int event);
380
381	int (*release)(struct inode *inode, struct file *file);
382
383	/*
384	 * register_event() callback will be used to add new userspace
385	 * waiter for changes related to the cftype. Implement it if
386	 * you want to provide this functionality. Use eventfd_signal()
387	 * on eventfd to send notification to userspace.
388	 */
389	int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
390			struct eventfd_ctx *eventfd, const char *args);
391	/*
392	 * unregister_event() callback will be called when userspace
393	 * closes the eventfd or on cgroup removing.
394	 * This callback must be implemented, if you want provide
395	 * notification functionality.
396	 */
397	void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
398			struct eventfd_ctx *eventfd);
399};
400
401struct cgroup_scanner {
402	struct cgroup *cg;
403	int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
404	void (*process_task)(struct task_struct *p,
405			struct cgroup_scanner *scan);
406	struct ptr_heap *heap;
407	void *data;
408};
409
410/*
411 * Add a new file to the given cgroup directory. Should only be
412 * called by subsystems from within a populate() method
413 */
414int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
415		       const struct cftype *cft);
416
417/*
418 * Add a set of new files to the given cgroup directory. Should
419 * only be called by subsystems from within a populate() method
420 */
421int cgroup_add_files(struct cgroup *cgrp,
422			struct cgroup_subsys *subsys,
423			const struct cftype cft[],
424			int count);
425
426int cgroup_is_removed(const struct cgroup *cgrp);
427
428int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
429
430int cgroup_task_count(const struct cgroup *cgrp);
431
432/* Return true if cgrp is a descendant of the task's cgroup */
433int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
434
435/*
436 * When the subsys has to access css and may add permanent refcnt to css,
437 * it should take care of racy conditions with rmdir(). Following set of
438 * functions, is for stop/restart rmdir if necessary.
439 * Because these will call css_get/put, "css" should be alive css.
440 *
441 *  cgroup_exclude_rmdir();
442 *  ...do some jobs which may access arbitrary empty cgroup
443 *  cgroup_release_and_wakeup_rmdir();
444 *
445 *  When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
446 *  it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
447 */
448
449void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
450void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
451
452/*
453 * Control Group subsystem type.
454 * See Documentation/cgroups/cgroups.txt for details
455 */
456
457struct cgroup_subsys {
458	struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
459						  struct cgroup *cgrp);
460	int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
461	void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
462	int (*allow_attach)(struct cgroup *cgrp, struct task_struct *tsk);
463	int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
464			  struct task_struct *tsk);
465	int (*can_attach_task)(struct cgroup *cgrp, struct task_struct *tsk);
466	void (*cancel_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
467			      struct task_struct *tsk);
468	void (*pre_attach)(struct cgroup *cgrp);
469	void (*attach_task)(struct cgroup *cgrp, struct task_struct *tsk);
470	void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
471		       struct cgroup *old_cgrp, struct task_struct *tsk);
472	void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
473	void (*exit)(struct cgroup_subsys *ss, struct cgroup *cgrp,
474			struct cgroup *old_cgrp, struct task_struct *task);
475	int (*populate)(struct cgroup_subsys *ss,
476			struct cgroup *cgrp);
477	void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
478	void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
479
480	int subsys_id;
481	int active;
482	int disabled;
483	int early_init;
484	/*
485	 * True if this subsys uses ID. ID is not available before cgroup_init()
486	 * (not available in early_init time.)
487	 */
488	bool use_id;
489#define MAX_CGROUP_TYPE_NAMELEN 32
490	const char *name;
491
492	/*
493	 * Protects sibling/children links of cgroups in this
494	 * hierarchy, plus protects which hierarchy (or none) the
495	 * subsystem is a part of (i.e. root/sibling).  To avoid
496	 * potential deadlocks, the following operations should not be
497	 * undertaken while holding any hierarchy_mutex:
498	 *
499	 * - allocating memory
500	 * - initiating hotplug events
501	 */
502	struct mutex hierarchy_mutex;
503	struct lock_class_key subsys_key;
504
505	/*
506	 * Link to parent, and list entry in parent's children.
507	 * Protected by this->hierarchy_mutex and cgroup_lock()
508	 */
509	struct cgroupfs_root *root;
510	struct list_head sibling;
511	/* used when use_id == true */
512	struct idr idr;
513	spinlock_t id_lock;
514
515	/* should be defined only by modular subsystems */
516	struct module *module;
517};
518
519#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
520#include <linux/cgroup_subsys.h>
521#undef SUBSYS
522
523static inline struct cgroup_subsys_state *cgroup_subsys_state(
524	struct cgroup *cgrp, int subsys_id)
525{
526	return cgrp->subsys[subsys_id];
527}
528
529/*
530 * function to get the cgroup_subsys_state which allows for extra
531 * rcu_dereference_check() conditions, such as locks used during the
532 * cgroup_subsys::attach() methods.
533 */
534#define task_subsys_state_check(task, subsys_id, __c)			\
535	rcu_dereference_check(task->cgroups->subsys[subsys_id],		\
536			      lockdep_is_held(&task->alloc_lock) ||	\
537			      cgroup_lock_is_held() || (__c))
538
539static inline struct cgroup_subsys_state *
540task_subsys_state(struct task_struct *task, int subsys_id)
541{
542	return task_subsys_state_check(task, subsys_id, false);
543}
544
545static inline struct cgroup* task_cgroup(struct task_struct *task,
546					       int subsys_id)
547{
548	return task_subsys_state(task, subsys_id)->cgroup;
549}
550
551/* A cgroup_iter should be treated as an opaque object */
552struct cgroup_iter {
553	struct list_head *cg_link;
554	struct list_head *task;
555};
556
557/*
558 * To iterate across the tasks in a cgroup:
559 *
560 * 1) call cgroup_iter_start to initialize an iterator
561 *
562 * 2) call cgroup_iter_next() to retrieve member tasks until it
563 *    returns NULL or until you want to end the iteration
564 *
565 * 3) call cgroup_iter_end() to destroy the iterator.
566 *
567 * Or, call cgroup_scan_tasks() to iterate through every task in a
568 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
569 * the test_task() callback, but not while calling the process_task()
570 * callback.
571 */
572void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
573struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
574					struct cgroup_iter *it);
575void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
576int cgroup_scan_tasks(struct cgroup_scanner *scan);
577int cgroup_attach_task(struct cgroup *, struct task_struct *);
578int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
579
580static inline int cgroup_attach_task_current_cg(struct task_struct *tsk)
581{
582	return cgroup_attach_task_all(current, tsk);
583}
584
585/*
586 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
587 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
588 * CSS ID is assigned at cgroup allocation (create) automatically
589 * and removed when subsys calls free_css_id() function. This is because
590 * the lifetime of cgroup_subsys_state is subsys's matter.
591 *
592 * Looking up and scanning function should be called under rcu_read_lock().
593 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
594 * But the css returned by this routine can be "not populated yet" or "being
595 * destroyed". The caller should check css and cgroup's status.
596 */
597
598/*
599 * Typically Called at ->destroy(), or somewhere the subsys frees
600 * cgroup_subsys_state.
601 */
602void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
603
604/* Find a cgroup_subsys_state which has given ID */
605
606struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
607
608/*
609 * Get a cgroup whose id is greater than or equal to id under tree of root.
610 * Returning a cgroup_subsys_state or NULL.
611 */
612struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
613		struct cgroup_subsys_state *root, int *foundid);
614
615/* Returns true if root is ancestor of cg */
616bool css_is_ancestor(struct cgroup_subsys_state *cg,
617		     const struct cgroup_subsys_state *root);
618
619/* Get id and depth of css */
620unsigned short css_id(struct cgroup_subsys_state *css);
621unsigned short css_depth(struct cgroup_subsys_state *css);
622struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
623
624#else /* !CONFIG_CGROUPS */
625
626static inline int cgroup_init_early(void) { return 0; }
627static inline int cgroup_init(void) { return 0; }
628static inline void cgroup_fork(struct task_struct *p) {}
629static inline void cgroup_fork_callbacks(struct task_struct *p) {}
630static inline void cgroup_post_fork(struct task_struct *p) {}
631static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
632
633static inline void cgroup_lock(void) {}
634static inline void cgroup_unlock(void) {}
635static inline int cgroupstats_build(struct cgroupstats *stats,
636					struct dentry *dentry)
637{
638	return -EINVAL;
639}
640
641/* No cgroups - nothing to do */
642static inline int cgroup_attach_task_all(struct task_struct *from,
643					 struct task_struct *t)
644{
645	return 0;
646}
647static inline int cgroup_attach_task_current_cg(struct task_struct *t)
648{
649	return 0;
650}
651
652#endif /* !CONFIG_CGROUPS */
653
654#endif /* _LINUX_CGROUP_H */