xref: /linux-6.15/include/linux/cgroup.h (revision 8fdff1dc)
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/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>
22 
23 #ifdef CONFIG_CGROUPS
24 
25 struct cgroupfs_root;
26 struct cgroup_subsys;
27 struct inode;
28 struct cgroup;
29 struct css_id;
30 
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);
44 
45 extern const struct file_operations proc_cgroup_operations;
46 
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>
52 	CGROUP_SUBSYS_COUNT,
53 };
54 #undef IS_SUBSYS_ENABLED
55 #undef SUBSYS
56 
57 /* Per-subsystem/per-cgroup state maintained by the system. */
58 struct cgroup_subsys_state {
59 	/*
60 	 * The cgroup that this subsystem is attached to. Useful
61 	 * for subsystems that want to know about the cgroup
62 	 * hierarchy structure
63 	 */
64 	struct cgroup *cgroup;
65 
66 	/*
67 	 * State maintained by the cgroup system to allow subsystems
68 	 * to be "busy". Should be accessed via css_get(),
69 	 * css_tryget() and css_put().
70 	 */
71 
72 	atomic_t refcnt;
73 
74 	unsigned long flags;
75 	/* ID for this css, if possible */
76 	struct css_id __rcu *id;
77 
78 	/* Used to put @cgroup->dentry on the last css_put() */
79 	struct work_struct dput_work;
80 };
81 
82 /* bits in struct cgroup_subsys_state flags field */
83 enum {
84 	CSS_ROOT	= (1 << 0), /* this CSS is the root of the subsystem */
85 	CSS_ONLINE	= (1 << 1), /* between ->css_online() and ->css_offline() */
86 };
87 
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)
90 {
91 	atomic_add(count, &css->refcnt);
92 }
93 
94 /*
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
99  */
100 
101 static inline void css_get(struct cgroup_subsys_state *css)
102 {
103 	/* We don't need to reference count the root state */
104 	if (!(css->flags & CSS_ROOT))
105 		__css_get(css, 1);
106 }
107 
108 /*
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.
112  */
113 
114 extern bool __css_tryget(struct cgroup_subsys_state *css);
115 static inline bool css_tryget(struct cgroup_subsys_state *css)
116 {
117 	if (css->flags & CSS_ROOT)
118 		return true;
119 	return __css_tryget(css);
120 }
121 
122 /*
123  * css_put() should be called to release a reference taken by
124  * css_get() or css_tryget()
125  */
126 
127 extern void __css_put(struct cgroup_subsys_state *css);
128 static inline void css_put(struct cgroup_subsys_state *css)
129 {
130 	if (!(css->flags & CSS_ROOT))
131 		__css_put(css);
132 }
133 
134 /* bits in struct cgroup flags field */
135 enum {
136 	/* Control Group is dead */
137 	CGRP_REMOVED,
138 	/*
139 	 * Control Group has previously had a child cgroup or a task,
140 	 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
141 	 */
142 	CGRP_RELEASABLE,
143 	/* Control Group requires release notifications to userspace */
144 	CGRP_NOTIFY_ON_RELEASE,
145 	/*
146 	 * Clone the parent's configuration when creating a new child
147 	 * cpuset cgroup.  For historical reasons, this option can be
148 	 * specified at mount time and thus is implemented here.
149 	 */
150 	CGRP_CPUSET_CLONE_CHILDREN,
151 };
152 
153 struct cgroup {
154 	unsigned long flags;		/* "unsigned long" so bitops work */
155 
156 	/*
157 	 * count users of this cgroup. >0 means busy, but doesn't
158 	 * necessarily indicate the number of tasks in the cgroup
159 	 */
160 	atomic_t count;
161 
162 	int id;				/* ida allocated in-hierarchy ID */
163 
164 	/*
165 	 * We link our 'sibling' struct into our parent's 'children'.
166 	 * Our children link their 'sibling' into our 'children'.
167 	 */
168 	struct list_head sibling;	/* my parent's children */
169 	struct list_head children;	/* my children */
170 	struct list_head files;		/* my files */
171 
172 	struct cgroup *parent;		/* my parent */
173 	struct dentry *dentry;		/* cgroup fs entry, RCU protected */
174 
175 	/* Private pointers for each registered subsystem */
176 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
177 
178 	struct cgroupfs_root *root;
179 	struct cgroup *top_cgroup;
180 
181 	/*
182 	 * List of cg_cgroup_links pointing at css_sets with
183 	 * tasks in this cgroup. Protected by css_set_lock
184 	 */
185 	struct list_head css_sets;
186 
187 	struct list_head allcg_node;	/* cgroupfs_root->allcg_list */
188 	struct list_head cft_q_node;	/* used during cftype add/rm */
189 
190 	/*
191 	 * Linked list running through all cgroups that can
192 	 * potentially be reaped by the release agent. Protected by
193 	 * release_list_lock
194 	 */
195 	struct list_head release_list;
196 
197 	/*
198 	 * list of pidlists, up to two for each namespace (one for procs, one
199 	 * for tasks); created on demand.
200 	 */
201 	struct list_head pidlists;
202 	struct mutex pidlist_mutex;
203 
204 	/* For RCU-protected deletion */
205 	struct rcu_head rcu_head;
206 
207 	/* List of events which userspace want to receive */
208 	struct list_head event_list;
209 	spinlock_t event_list_lock;
210 
211 	/* directory xattrs */
212 	struct simple_xattrs xattrs;
213 };
214 
215 /*
216  * A css_set is a structure holding pointers to a set of
217  * cgroup_subsys_state objects. This saves space in the task struct
218  * object and speeds up fork()/exit(), since a single inc/dec and a
219  * list_add()/del() can bump the reference count on the entire cgroup
220  * set for a task.
221  */
222 
223 struct css_set {
224 
225 	/* Reference count */
226 	atomic_t refcount;
227 
228 	/*
229 	 * List running through all cgroup groups in the same hash
230 	 * slot. Protected by css_set_lock
231 	 */
232 	struct hlist_node hlist;
233 
234 	/*
235 	 * List running through all tasks using this cgroup
236 	 * group. Protected by css_set_lock
237 	 */
238 	struct list_head tasks;
239 
240 	/*
241 	 * List of cg_cgroup_link objects on link chains from
242 	 * cgroups referenced from this css_set. Protected by
243 	 * css_set_lock
244 	 */
245 	struct list_head cg_links;
246 
247 	/*
248 	 * Set of subsystem states, one for each subsystem. This array
249 	 * is immutable after creation apart from the init_css_set
250 	 * during subsystem registration (at boot time) and modular subsystem
251 	 * loading/unloading.
252 	 */
253 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
254 
255 	/* For RCU-protected deletion */
256 	struct rcu_head rcu_head;
257 };
258 
259 /*
260  * cgroup_map_cb is an abstract callback API for reporting map-valued
261  * control files
262  */
263 
264 struct cgroup_map_cb {
265 	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
266 	void *state;
267 };
268 
269 /*
270  * struct cftype: handler definitions for cgroup control files
271  *
272  * When reading/writing to a file:
273  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
274  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
275  */
276 
277 /* cftype->flags */
278 #define CFTYPE_ONLY_ON_ROOT	(1U << 0)	/* only create on root cg */
279 #define CFTYPE_NOT_ON_ROOT	(1U << 1)	/* don't create on root cg */
280 
281 #define MAX_CFTYPE_NAME		64
282 
283 struct cftype {
284 	/*
285 	 * By convention, the name should begin with the name of the
286 	 * subsystem, followed by a period.  Zero length string indicates
287 	 * end of cftype array.
288 	 */
289 	char name[MAX_CFTYPE_NAME];
290 	int private;
291 	/*
292 	 * If not 0, file mode is set to this value, otherwise it will
293 	 * be figured out automatically
294 	 */
295 	umode_t mode;
296 
297 	/*
298 	 * If non-zero, defines the maximum length of string that can
299 	 * be passed to write_string; defaults to 64
300 	 */
301 	size_t max_write_len;
302 
303 	/* CFTYPE_* flags */
304 	unsigned int flags;
305 
306 	/* file xattrs */
307 	struct simple_xattrs xattrs;
308 
309 	int (*open)(struct inode *inode, struct file *file);
310 	ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
311 			struct file *file,
312 			char __user *buf, size_t nbytes, loff_t *ppos);
313 	/*
314 	 * read_u64() is a shortcut for the common case of returning a
315 	 * single integer. Use it in place of read()
316 	 */
317 	u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
318 	/*
319 	 * read_s64() is a signed version of read_u64()
320 	 */
321 	s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
322 	/*
323 	 * read_map() is used for defining a map of key/value
324 	 * pairs. It should call cb->fill(cb, key, value) for each
325 	 * entry. The key/value pairs (and their ordering) should not
326 	 * change between reboots.
327 	 */
328 	int (*read_map)(struct cgroup *cont, struct cftype *cft,
329 			struct cgroup_map_cb *cb);
330 	/*
331 	 * read_seq_string() is used for outputting a simple sequence
332 	 * using seqfile.
333 	 */
334 	int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
335 			       struct seq_file *m);
336 
337 	ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
338 			 struct file *file,
339 			 const char __user *buf, size_t nbytes, loff_t *ppos);
340 
341 	/*
342 	 * write_u64() is a shortcut for the common case of accepting
343 	 * a single integer (as parsed by simple_strtoull) from
344 	 * userspace. Use in place of write(); return 0 or error.
345 	 */
346 	int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
347 	/*
348 	 * write_s64() is a signed version of write_u64()
349 	 */
350 	int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
351 
352 	/*
353 	 * write_string() is passed a nul-terminated kernelspace
354 	 * buffer of maximum length determined by max_write_len.
355 	 * Returns 0 or -ve error code.
356 	 */
357 	int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
358 			    const char *buffer);
359 	/*
360 	 * trigger() callback can be used to get some kick from the
361 	 * userspace, when the actual string written is not important
362 	 * at all. The private field can be used to determine the
363 	 * kick type for multiplexing.
364 	 */
365 	int (*trigger)(struct cgroup *cgrp, unsigned int event);
366 
367 	int (*release)(struct inode *inode, struct file *file);
368 
369 	/*
370 	 * register_event() callback will be used to add new userspace
371 	 * waiter for changes related to the cftype. Implement it if
372 	 * you want to provide this functionality. Use eventfd_signal()
373 	 * on eventfd to send notification to userspace.
374 	 */
375 	int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
376 			struct eventfd_ctx *eventfd, const char *args);
377 	/*
378 	 * unregister_event() callback will be called when userspace
379 	 * closes the eventfd or on cgroup removing.
380 	 * This callback must be implemented, if you want provide
381 	 * notification functionality.
382 	 */
383 	void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
384 			struct eventfd_ctx *eventfd);
385 };
386 
387 /*
388  * cftype_sets describe cftypes belonging to a subsystem and are chained at
389  * cgroup_subsys->cftsets.  Each cftset points to an array of cftypes
390  * terminated by zero length name.
391  */
392 struct cftype_set {
393 	struct list_head		node;	/* chained at subsys->cftsets */
394 	struct cftype			*cfts;
395 };
396 
397 struct cgroup_scanner {
398 	struct cgroup *cg;
399 	int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
400 	void (*process_task)(struct task_struct *p,
401 			struct cgroup_scanner *scan);
402 	struct ptr_heap *heap;
403 	void *data;
404 };
405 
406 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
407 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
408 
409 int cgroup_is_removed(const struct cgroup *cgrp);
410 
411 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
412 
413 int cgroup_task_count(const struct cgroup *cgrp);
414 
415 /* Return true if cgrp is a descendant of the task's cgroup */
416 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
417 
418 /*
419  * Control Group taskset, used to pass around set of tasks to cgroup_subsys
420  * methods.
421  */
422 struct cgroup_taskset;
423 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
424 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
425 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
426 int cgroup_taskset_size(struct cgroup_taskset *tset);
427 
428 /**
429  * cgroup_taskset_for_each - iterate cgroup_taskset
430  * @task: the loop cursor
431  * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
432  * @tset: taskset to iterate
433  */
434 #define cgroup_taskset_for_each(task, skip_cgrp, tset)			\
435 	for ((task) = cgroup_taskset_first((tset)); (task);		\
436 	     (task) = cgroup_taskset_next((tset)))			\
437 		if (!(skip_cgrp) ||					\
438 		    cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
439 
440 /*
441  * Control Group subsystem type.
442  * See Documentation/cgroups/cgroups.txt for details
443  */
444 
445 struct cgroup_subsys {
446 	struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
447 	int (*css_online)(struct cgroup *cgrp);
448 	void (*css_offline)(struct cgroup *cgrp);
449 	void (*css_free)(struct cgroup *cgrp);
450 
451 	int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
452 	void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
453 	void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
454 	void (*fork)(struct task_struct *task);
455 	void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
456 		     struct task_struct *task);
457 	void (*bind)(struct cgroup *root);
458 
459 	int subsys_id;
460 	int active;
461 	int disabled;
462 	int early_init;
463 	/*
464 	 * True if this subsys uses ID. ID is not available before cgroup_init()
465 	 * (not available in early_init time.)
466 	 */
467 	bool use_id;
468 
469 	/*
470 	 * If %false, this subsystem is properly hierarchical -
471 	 * configuration, resource accounting and restriction on a parent
472 	 * cgroup cover those of its children.  If %true, hierarchy support
473 	 * is broken in some ways - some subsystems ignore hierarchy
474 	 * completely while others are only implemented half-way.
475 	 *
476 	 * It's now disallowed to create nested cgroups if the subsystem is
477 	 * broken and cgroup core will emit a warning message on such
478 	 * cases.  Eventually, all subsystems will be made properly
479 	 * hierarchical and this will go away.
480 	 */
481 	bool broken_hierarchy;
482 	bool warned_broken_hierarchy;
483 
484 #define MAX_CGROUP_TYPE_NAMELEN 32
485 	const char *name;
486 
487 	/*
488 	 * Link to parent, and list entry in parent's children.
489 	 * Protected by cgroup_lock()
490 	 */
491 	struct cgroupfs_root *root;
492 	struct list_head sibling;
493 	/* used when use_id == true */
494 	struct idr idr;
495 	spinlock_t id_lock;
496 
497 	/* list of cftype_sets */
498 	struct list_head cftsets;
499 
500 	/* base cftypes, automatically [de]registered with subsys itself */
501 	struct cftype *base_cftypes;
502 	struct cftype_set base_cftset;
503 
504 	/* should be defined only by modular subsystems */
505 	struct module *module;
506 };
507 
508 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
509 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
510 #include <linux/cgroup_subsys.h>
511 #undef IS_SUBSYS_ENABLED
512 #undef SUBSYS
513 
514 static inline struct cgroup_subsys_state *cgroup_subsys_state(
515 	struct cgroup *cgrp, int subsys_id)
516 {
517 	return cgrp->subsys[subsys_id];
518 }
519 
520 /*
521  * function to get the cgroup_subsys_state which allows for extra
522  * rcu_dereference_check() conditions, such as locks used during the
523  * cgroup_subsys::attach() methods.
524  */
525 #define task_subsys_state_check(task, subsys_id, __c)			\
526 	rcu_dereference_check(task->cgroups->subsys[subsys_id],		\
527 			      lockdep_is_held(&task->alloc_lock) ||	\
528 			      cgroup_lock_is_held() || (__c))
529 
530 static inline struct cgroup_subsys_state *
531 task_subsys_state(struct task_struct *task, int subsys_id)
532 {
533 	return task_subsys_state_check(task, subsys_id, false);
534 }
535 
536 static inline struct cgroup* task_cgroup(struct task_struct *task,
537 					       int subsys_id)
538 {
539 	return task_subsys_state(task, subsys_id)->cgroup;
540 }
541 
542 /**
543  * cgroup_for_each_child - iterate through children of a cgroup
544  * @pos: the cgroup * to use as the loop cursor
545  * @cgroup: cgroup whose children to walk
546  *
547  * Walk @cgroup's children.  Must be called under rcu_read_lock().  A child
548  * cgroup which hasn't finished ->css_online() or already has finished
549  * ->css_offline() may show up during traversal and it's each subsystem's
550  * responsibility to verify that each @pos is alive.
551  *
552  * If a subsystem synchronizes against the parent in its ->css_online() and
553  * before starting iterating, a cgroup which finished ->css_online() is
554  * guaranteed to be visible in the future iterations.
555  */
556 #define cgroup_for_each_child(pos, cgroup)				\
557 	list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
558 
559 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
560 					  struct cgroup *cgroup);
561 
562 /**
563  * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
564  * @pos: the cgroup * to use as the loop cursor
565  * @cgroup: cgroup whose descendants to walk
566  *
567  * Walk @cgroup's descendants.  Must be called under rcu_read_lock().  A
568  * descendant cgroup which hasn't finished ->css_online() or already has
569  * finished ->css_offline() may show up during traversal and it's each
570  * subsystem's responsibility to verify that each @pos is alive.
571  *
572  * If a subsystem synchronizes against the parent in its ->css_online() and
573  * before starting iterating, and synchronizes against @pos on each
574  * iteration, any descendant cgroup which finished ->css_offline() is
575  * guaranteed to be visible in the future iterations.
576  *
577  * In other words, the following guarantees that a descendant can't escape
578  * state updates of its ancestors.
579  *
580  * my_online(@cgrp)
581  * {
582  *	Lock @cgrp->parent and @cgrp;
583  *	Inherit state from @cgrp->parent;
584  *	Unlock both.
585  * }
586  *
587  * my_update_state(@cgrp)
588  * {
589  *	Lock @cgrp;
590  *	Update @cgrp's state;
591  *	Unlock @cgrp;
592  *
593  *	cgroup_for_each_descendant_pre(@pos, @cgrp) {
594  *		Lock @pos;
595  *		Verify @pos is alive and inherit state from @pos->parent;
596  *		Unlock @pos;
597  *	}
598  * }
599  *
600  * As long as the inheriting step, including checking the parent state, is
601  * enclosed inside @pos locking, double-locking the parent isn't necessary
602  * while inheriting.  The state update to the parent is guaranteed to be
603  * visible by walking order and, as long as inheriting operations to the
604  * same @pos are atomic to each other, multiple updates racing each other
605  * still result in the correct state.  It's guaranateed that at least one
606  * inheritance happens for any cgroup after the latest update to its
607  * parent.
608  *
609  * If checking parent's state requires locking the parent, each inheriting
610  * iteration should lock and unlock both @pos->parent and @pos.
611  *
612  * Alternatively, a subsystem may choose to use a single global lock to
613  * synchronize ->css_online() and ->css_offline() against tree-walking
614  * operations.
615  */
616 #define cgroup_for_each_descendant_pre(pos, cgroup)			\
617 	for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos);	\
618 	     pos = cgroup_next_descendant_pre((pos), (cgroup)))
619 
620 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
621 					   struct cgroup *cgroup);
622 
623 /**
624  * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
625  * @pos: the cgroup * to use as the loop cursor
626  * @cgroup: cgroup whose descendants to walk
627  *
628  * Similar to cgroup_for_each_descendant_pre() but performs post-order
629  * traversal instead.  Note that the walk visibility guarantee described in
630  * pre-order walk doesn't apply the same to post-order walks.
631  */
632 #define cgroup_for_each_descendant_post(pos, cgroup)			\
633 	for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos);	\
634 	     pos = cgroup_next_descendant_post((pos), (cgroup)))
635 
636 /* A cgroup_iter should be treated as an opaque object */
637 struct cgroup_iter {
638 	struct list_head *cg_link;
639 	struct list_head *task;
640 };
641 
642 /*
643  * To iterate across the tasks in a cgroup:
644  *
645  * 1) call cgroup_iter_start to initialize an iterator
646  *
647  * 2) call cgroup_iter_next() to retrieve member tasks until it
648  *    returns NULL or until you want to end the iteration
649  *
650  * 3) call cgroup_iter_end() to destroy the iterator.
651  *
652  * Or, call cgroup_scan_tasks() to iterate through every task in a
653  * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
654  * the test_task() callback, but not while calling the process_task()
655  * callback.
656  */
657 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
658 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
659 					struct cgroup_iter *it);
660 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
661 int cgroup_scan_tasks(struct cgroup_scanner *scan);
662 int cgroup_attach_task(struct cgroup *, struct task_struct *);
663 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
664 
665 /*
666  * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
667  * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
668  * CSS ID is assigned at cgroup allocation (create) automatically
669  * and removed when subsys calls free_css_id() function. This is because
670  * the lifetime of cgroup_subsys_state is subsys's matter.
671  *
672  * Looking up and scanning function should be called under rcu_read_lock().
673  * Taking cgroup_mutex is not necessary for following calls.
674  * But the css returned by this routine can be "not populated yet" or "being
675  * destroyed". The caller should check css and cgroup's status.
676  */
677 
678 /*
679  * Typically Called at ->destroy(), or somewhere the subsys frees
680  * cgroup_subsys_state.
681  */
682 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
683 
684 /* Find a cgroup_subsys_state which has given ID */
685 
686 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
687 
688 /*
689  * Get a cgroup whose id is greater than or equal to id under tree of root.
690  * Returning a cgroup_subsys_state or NULL.
691  */
692 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
693 		struct cgroup_subsys_state *root, int *foundid);
694 
695 /* Returns true if root is ancestor of cg */
696 bool css_is_ancestor(struct cgroup_subsys_state *cg,
697 		     const struct cgroup_subsys_state *root);
698 
699 /* Get id and depth of css */
700 unsigned short css_id(struct cgroup_subsys_state *css);
701 unsigned short css_depth(struct cgroup_subsys_state *css);
702 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
703 
704 #else /* !CONFIG_CGROUPS */
705 
706 static inline int cgroup_init_early(void) { return 0; }
707 static inline int cgroup_init(void) { return 0; }
708 static inline void cgroup_fork(struct task_struct *p) {}
709 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
710 static inline void cgroup_post_fork(struct task_struct *p) {}
711 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
712 
713 static inline void cgroup_lock(void) {}
714 static inline void cgroup_unlock(void) {}
715 static inline int cgroupstats_build(struct cgroupstats *stats,
716 					struct dentry *dentry)
717 {
718 	return -EINVAL;
719 }
720 
721 /* No cgroups - nothing to do */
722 static inline int cgroup_attach_task_all(struct task_struct *from,
723 					 struct task_struct *t)
724 {
725 	return 0;
726 }
727 
728 #endif /* !CONFIG_CGROUPS */
729 
730 #endif /* _LINUX_CGROUP_H */
731