xref: /openbmc/linux/include/linux/cgroup.h (revision 7051924f771722c6dd235e693742cda6488ac700)
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/rwsem.h>
18 #include <linux/idr.h>
19 #include <linux/workqueue.h>
20 #include <linux/fs.h>
21 #include <linux/percpu-refcount.h>
22 #include <linux/seq_file.h>
23 #include <linux/kernfs.h>
24 #include <linux/wait.h>
25 
26 #ifdef CONFIG_CGROUPS
27 
28 struct cgroup_root;
29 struct cgroup_subsys;
30 struct inode;
31 struct cgroup;
32 
33 extern int cgroup_init_early(void);
34 extern int cgroup_init(void);
35 extern void cgroup_fork(struct task_struct *p);
36 extern void cgroup_post_fork(struct task_struct *p);
37 extern void cgroup_exit(struct task_struct *p);
38 extern int cgroupstats_build(struct cgroupstats *stats,
39 				struct dentry *dentry);
40 
41 extern int proc_cgroup_show(struct seq_file *, void *);
42 
43 /* define the enumeration of all cgroup subsystems */
44 #define SUBSYS(_x) _x ## _cgrp_id,
45 enum cgroup_subsys_id {
46 #include <linux/cgroup_subsys.h>
47 	CGROUP_SUBSYS_COUNT,
48 };
49 #undef SUBSYS
50 
51 /*
52  * Per-subsystem/per-cgroup state maintained by the system.  This is the
53  * fundamental structural building block that controllers deal with.
54  *
55  * Fields marked with "PI:" are public and immutable and may be accessed
56  * directly without synchronization.
57  */
58 struct cgroup_subsys_state {
59 	/* PI: the cgroup that this css is attached to */
60 	struct cgroup *cgroup;
61 
62 	/* PI: the cgroup subsystem that this css is attached to */
63 	struct cgroup_subsys *ss;
64 
65 	/* reference count - access via css_[try]get() and css_put() */
66 	struct percpu_ref refcnt;
67 
68 	/* PI: the parent css */
69 	struct cgroup_subsys_state *parent;
70 
71 	/* siblings list anchored at the parent's ->children */
72 	struct list_head sibling;
73 	struct list_head children;
74 
75 	/*
76 	 * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
77 	 * matching css can be looked up using css_from_id().
78 	 */
79 	int id;
80 
81 	unsigned int flags;
82 
83 	/*
84 	 * Monotonically increasing unique serial number which defines a
85 	 * uniform order among all csses.  It's guaranteed that all
86 	 * ->children lists are in the ascending order of ->serial_nr and
87 	 * used to allow interrupting and resuming iterations.
88 	 */
89 	u64 serial_nr;
90 
91 	/* percpu_ref killing and RCU release */
92 	struct rcu_head rcu_head;
93 	struct work_struct destroy_work;
94 };
95 
96 /* bits in struct cgroup_subsys_state flags field */
97 enum {
98 	CSS_NO_REF	= (1 << 0), /* no reference counting for this css */
99 	CSS_ONLINE	= (1 << 1), /* between ->css_online() and ->css_offline() */
100 	CSS_RELEASED	= (1 << 2), /* refcnt reached zero, released */
101 };
102 
103 /**
104  * css_get - obtain a reference on the specified css
105  * @css: target css
106  *
107  * The caller must already have a reference.
108  */
109 static inline void css_get(struct cgroup_subsys_state *css)
110 {
111 	if (!(css->flags & CSS_NO_REF))
112 		percpu_ref_get(&css->refcnt);
113 }
114 
115 /**
116  * css_tryget - try to obtain a reference on the specified css
117  * @css: target css
118  *
119  * Obtain a reference on @css unless it already has reached zero and is
120  * being released.  This function doesn't care whether @css is on or
121  * offline.  The caller naturally needs to ensure that @css is accessible
122  * but doesn't have to be holding a reference on it - IOW, RCU protected
123  * access is good enough for this function.  Returns %true if a reference
124  * count was successfully obtained; %false otherwise.
125  */
126 static inline bool css_tryget(struct cgroup_subsys_state *css)
127 {
128 	if (!(css->flags & CSS_NO_REF))
129 		return percpu_ref_tryget(&css->refcnt);
130 	return true;
131 }
132 
133 /**
134  * css_tryget_online - try to obtain a reference on the specified css if online
135  * @css: target css
136  *
137  * Obtain a reference on @css if it's online.  The caller naturally needs
138  * to ensure that @css is accessible but doesn't have to be holding a
139  * reference on it - IOW, RCU protected access is good enough for this
140  * function.  Returns %true if a reference count was successfully obtained;
141  * %false otherwise.
142  */
143 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
144 {
145 	if (!(css->flags & CSS_NO_REF))
146 		return percpu_ref_tryget_live(&css->refcnt);
147 	return true;
148 }
149 
150 /**
151  * css_put - put a css reference
152  * @css: target css
153  *
154  * Put a reference obtained via css_get() and css_tryget_online().
155  */
156 static inline void css_put(struct cgroup_subsys_state *css)
157 {
158 	if (!(css->flags & CSS_NO_REF))
159 		percpu_ref_put(&css->refcnt);
160 }
161 
162 /* bits in struct cgroup flags field */
163 enum {
164 	/*
165 	 * Control Group has previously had a child cgroup or a task,
166 	 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
167 	 */
168 	CGRP_RELEASABLE,
169 	/* Control Group requires release notifications to userspace */
170 	CGRP_NOTIFY_ON_RELEASE,
171 	/*
172 	 * Clone the parent's configuration when creating a new child
173 	 * cpuset cgroup.  For historical reasons, this option can be
174 	 * specified at mount time and thus is implemented here.
175 	 */
176 	CGRP_CPUSET_CLONE_CHILDREN,
177 };
178 
179 struct cgroup {
180 	/* self css with NULL ->ss, points back to this cgroup */
181 	struct cgroup_subsys_state self;
182 
183 	unsigned long flags;		/* "unsigned long" so bitops work */
184 
185 	/*
186 	 * idr allocated in-hierarchy ID.
187 	 *
188 	 * ID 0 is not used, the ID of the root cgroup is always 1, and a
189 	 * new cgroup will be assigned with a smallest available ID.
190 	 *
191 	 * Allocating/Removing ID must be protected by cgroup_mutex.
192 	 */
193 	int id;
194 
195 	/*
196 	 * If this cgroup contains any tasks, it contributes one to
197 	 * populated_cnt.  All children with non-zero popuplated_cnt of
198 	 * their own contribute one.  The count is zero iff there's no task
199 	 * in this cgroup or its subtree.
200 	 */
201 	int populated_cnt;
202 
203 	struct kernfs_node *kn;		/* cgroup kernfs entry */
204 	struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
205 
206 	/*
207 	 * The bitmask of subsystems enabled on the child cgroups.
208 	 * ->subtree_control is the one configured through
209 	 * "cgroup.subtree_control" while ->child_subsys_mask is the
210 	 * effective one which may have more subsystems enabled.
211 	 * Controller knobs are made available iff it's enabled in
212 	 * ->subtree_control.
213 	 */
214 	unsigned int subtree_control;
215 	unsigned int child_subsys_mask;
216 
217 	/* Private pointers for each registered subsystem */
218 	struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
219 
220 	struct cgroup_root *root;
221 
222 	/*
223 	 * List of cgrp_cset_links pointing at css_sets with tasks in this
224 	 * cgroup.  Protected by css_set_lock.
225 	 */
226 	struct list_head cset_links;
227 
228 	/*
229 	 * On the default hierarchy, a css_set for a cgroup with some
230 	 * susbsys disabled will point to css's which are associated with
231 	 * the closest ancestor which has the subsys enabled.  The
232 	 * following lists all css_sets which point to this cgroup's css
233 	 * for the given subsystem.
234 	 */
235 	struct list_head e_csets[CGROUP_SUBSYS_COUNT];
236 
237 	/*
238 	 * Linked list running through all cgroups that can
239 	 * potentially be reaped by the release agent. Protected by
240 	 * release_list_lock
241 	 */
242 	struct list_head release_list;
243 
244 	/*
245 	 * list of pidlists, up to two for each namespace (one for procs, one
246 	 * for tasks); created on demand.
247 	 */
248 	struct list_head pidlists;
249 	struct mutex pidlist_mutex;
250 
251 	/* used to wait for offlining of csses */
252 	wait_queue_head_t offline_waitq;
253 };
254 
255 #define MAX_CGROUP_ROOT_NAMELEN 64
256 
257 /* cgroup_root->flags */
258 enum {
259 	CGRP_ROOT_SANE_BEHAVIOR	= (1 << 0), /* __DEVEL__sane_behavior specified */
260 	CGRP_ROOT_NOPREFIX	= (1 << 1), /* mounted subsystems have no named prefix */
261 	CGRP_ROOT_XATTR		= (1 << 2), /* supports extended attributes */
262 };
263 
264 /*
265  * A cgroup_root represents the root of a cgroup hierarchy, and may be
266  * associated with a kernfs_root to form an active hierarchy.  This is
267  * internal to cgroup core.  Don't access directly from controllers.
268  */
269 struct cgroup_root {
270 	struct kernfs_root *kf_root;
271 
272 	/* The bitmask of subsystems attached to this hierarchy */
273 	unsigned int subsys_mask;
274 
275 	/* Unique id for this hierarchy. */
276 	int hierarchy_id;
277 
278 	/* The root cgroup.  Root is destroyed on its release. */
279 	struct cgroup cgrp;
280 
281 	/* Number of cgroups in the hierarchy, used only for /proc/cgroups */
282 	atomic_t nr_cgrps;
283 
284 	/* A list running through the active hierarchies */
285 	struct list_head root_list;
286 
287 	/* Hierarchy-specific flags */
288 	unsigned int flags;
289 
290 	/* IDs for cgroups in this hierarchy */
291 	struct idr cgroup_idr;
292 
293 	/* The path to use for release notifications. */
294 	char release_agent_path[PATH_MAX];
295 
296 	/* The name for this hierarchy - may be empty */
297 	char name[MAX_CGROUP_ROOT_NAMELEN];
298 };
299 
300 /*
301  * A css_set is a structure holding pointers to a set of
302  * cgroup_subsys_state objects. This saves space in the task struct
303  * object and speeds up fork()/exit(), since a single inc/dec and a
304  * list_add()/del() can bump the reference count on the entire cgroup
305  * set for a task.
306  */
307 
308 struct css_set {
309 
310 	/* Reference count */
311 	atomic_t refcount;
312 
313 	/*
314 	 * List running through all cgroup groups in the same hash
315 	 * slot. Protected by css_set_lock
316 	 */
317 	struct hlist_node hlist;
318 
319 	/*
320 	 * Lists running through all tasks using this cgroup group.
321 	 * mg_tasks lists tasks which belong to this cset but are in the
322 	 * process of being migrated out or in.  Protected by
323 	 * css_set_rwsem, but, during migration, once tasks are moved to
324 	 * mg_tasks, it can be read safely while holding cgroup_mutex.
325 	 */
326 	struct list_head tasks;
327 	struct list_head mg_tasks;
328 
329 	/*
330 	 * List of cgrp_cset_links pointing at cgroups referenced from this
331 	 * css_set.  Protected by css_set_lock.
332 	 */
333 	struct list_head cgrp_links;
334 
335 	/* the default cgroup associated with this css_set */
336 	struct cgroup *dfl_cgrp;
337 
338 	/*
339 	 * Set of subsystem states, one for each subsystem. This array is
340 	 * immutable after creation apart from the init_css_set during
341 	 * subsystem registration (at boot time).
342 	 */
343 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
344 
345 	/*
346 	 * List of csets participating in the on-going migration either as
347 	 * source or destination.  Protected by cgroup_mutex.
348 	 */
349 	struct list_head mg_preload_node;
350 	struct list_head mg_node;
351 
352 	/*
353 	 * If this cset is acting as the source of migration the following
354 	 * two fields are set.  mg_src_cgrp is the source cgroup of the
355 	 * on-going migration and mg_dst_cset is the destination cset the
356 	 * target tasks on this cset should be migrated to.  Protected by
357 	 * cgroup_mutex.
358 	 */
359 	struct cgroup *mg_src_cgrp;
360 	struct css_set *mg_dst_cset;
361 
362 	/*
363 	 * On the default hierarhcy, ->subsys[ssid] may point to a css
364 	 * attached to an ancestor instead of the cgroup this css_set is
365 	 * associated with.  The following node is anchored at
366 	 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
367 	 * iterate through all css's attached to a given cgroup.
368 	 */
369 	struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
370 
371 	/* For RCU-protected deletion */
372 	struct rcu_head rcu_head;
373 };
374 
375 /*
376  * struct cftype: handler definitions for cgroup control files
377  *
378  * When reading/writing to a file:
379  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
380  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
381  */
382 
383 /* cftype->flags */
384 enum {
385 	CFTYPE_ONLY_ON_ROOT	= (1 << 0),	/* only create on root cgrp */
386 	CFTYPE_NOT_ON_ROOT	= (1 << 1),	/* don't create on root cgrp */
387 	CFTYPE_NO_PREFIX	= (1 << 3),	/* (DON'T USE FOR NEW FILES) no subsys prefix */
388 
389 	/* internal flags, do not use outside cgroup core proper */
390 	__CFTYPE_ONLY_ON_DFL	= (1 << 16),	/* only on default hierarchy */
391 	__CFTYPE_NOT_ON_DFL	= (1 << 17),	/* not on default hierarchy */
392 };
393 
394 #define MAX_CFTYPE_NAME		64
395 
396 struct cftype {
397 	/*
398 	 * By convention, the name should begin with the name of the
399 	 * subsystem, followed by a period.  Zero length string indicates
400 	 * end of cftype array.
401 	 */
402 	char name[MAX_CFTYPE_NAME];
403 	int private;
404 	/*
405 	 * If not 0, file mode is set to this value, otherwise it will
406 	 * be figured out automatically
407 	 */
408 	umode_t mode;
409 
410 	/*
411 	 * The maximum length of string, excluding trailing nul, that can
412 	 * be passed to write.  If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
413 	 */
414 	size_t max_write_len;
415 
416 	/* CFTYPE_* flags */
417 	unsigned int flags;
418 
419 	/*
420 	 * Fields used for internal bookkeeping.  Initialized automatically
421 	 * during registration.
422 	 */
423 	struct cgroup_subsys *ss;	/* NULL for cgroup core files */
424 	struct list_head node;		/* anchored at ss->cfts */
425 	struct kernfs_ops *kf_ops;
426 
427 	/*
428 	 * read_u64() is a shortcut for the common case of returning a
429 	 * single integer. Use it in place of read()
430 	 */
431 	u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
432 	/*
433 	 * read_s64() is a signed version of read_u64()
434 	 */
435 	s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
436 
437 	/* generic seq_file read interface */
438 	int (*seq_show)(struct seq_file *sf, void *v);
439 
440 	/* optional ops, implement all or none */
441 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
442 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
443 	void (*seq_stop)(struct seq_file *sf, void *v);
444 
445 	/*
446 	 * write_u64() is a shortcut for the common case of accepting
447 	 * a single integer (as parsed by simple_strtoull) from
448 	 * userspace. Use in place of write(); return 0 or error.
449 	 */
450 	int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
451 			 u64 val);
452 	/*
453 	 * write_s64() is a signed version of write_u64()
454 	 */
455 	int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
456 			 s64 val);
457 
458 	/*
459 	 * write() is the generic write callback which maps directly to
460 	 * kernfs write operation and overrides all other operations.
461 	 * Maximum write size is determined by ->max_write_len.  Use
462 	 * of_css/cft() to access the associated css and cft.
463 	 */
464 	ssize_t (*write)(struct kernfs_open_file *of,
465 			 char *buf, size_t nbytes, loff_t off);
466 
467 #ifdef CONFIG_DEBUG_LOCK_ALLOC
468 	struct lock_class_key	lockdep_key;
469 #endif
470 };
471 
472 extern struct cgroup_root cgrp_dfl_root;
473 extern struct css_set init_css_set;
474 
475 /**
476  * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
477  * @cgrp: the cgroup of interest
478  *
479  * The default hierarchy is the v2 interface of cgroup and this function
480  * can be used to test whether a cgroup is on the default hierarchy for
481  * cases where a subsystem should behave differnetly depending on the
482  * interface version.
483  *
484  * The set of behaviors which change on the default hierarchy are still
485  * being determined and the mount option is prefixed with __DEVEL__.
486  *
487  * List of changed behaviors:
488  *
489  * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
490  *   and "name" are disallowed.
491  *
492  * - When mounting an existing superblock, mount options should match.
493  *
494  * - Remount is disallowed.
495  *
496  * - rename(2) is disallowed.
497  *
498  * - "tasks" is removed.  Everything should be at process granularity.  Use
499  *   "cgroup.procs" instead.
500  *
501  * - "cgroup.procs" is not sorted.  pids will be unique unless they got
502  *   recycled inbetween reads.
503  *
504  * - "release_agent" and "notify_on_release" are removed.  Replacement
505  *   notification mechanism will be implemented.
506  *
507  * - "cgroup.clone_children" is removed.
508  *
509  * - "cgroup.subtree_populated" is available.  Its value is 0 if the cgroup
510  *   and its descendants contain no task; otherwise, 1.  The file also
511  *   generates kernfs notification which can be monitored through poll and
512  *   [di]notify when the value of the file changes.
513  *
514  * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
515  *   take masks of ancestors with non-empty cpus/mems, instead of being
516  *   moved to an ancestor.
517  *
518  * - cpuset: a task can be moved into an empty cpuset, and again it takes
519  *   masks of ancestors.
520  *
521  * - memcg: use_hierarchy is on by default and the cgroup file for the flag
522  *   is not created.
523  *
524  * - blkcg: blk-throttle becomes properly hierarchical.
525  *
526  * - debug: disallowed on the default hierarchy.
527  */
528 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
529 {
530 	return cgrp->root == &cgrp_dfl_root;
531 }
532 
533 /* no synchronization, the result can only be used as a hint */
534 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
535 {
536 	return !list_empty(&cgrp->cset_links);
537 }
538 
539 /* returns ino associated with a cgroup, 0 indicates unmounted root */
540 static inline ino_t cgroup_ino(struct cgroup *cgrp)
541 {
542 	if (cgrp->kn)
543 		return cgrp->kn->ino;
544 	else
545 		return 0;
546 }
547 
548 /* cft/css accessors for cftype->write() operation */
549 static inline struct cftype *of_cft(struct kernfs_open_file *of)
550 {
551 	return of->kn->priv;
552 }
553 
554 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
555 
556 /* cft/css accessors for cftype->seq_*() operations */
557 static inline struct cftype *seq_cft(struct seq_file *seq)
558 {
559 	return of_cft(seq->private);
560 }
561 
562 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
563 {
564 	return of_css(seq->private);
565 }
566 
567 /*
568  * Name / path handling functions.  All are thin wrappers around the kernfs
569  * counterparts and can be called under any context.
570  */
571 
572 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
573 {
574 	return kernfs_name(cgrp->kn, buf, buflen);
575 }
576 
577 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
578 					      size_t buflen)
579 {
580 	return kernfs_path(cgrp->kn, buf, buflen);
581 }
582 
583 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
584 {
585 	pr_cont_kernfs_name(cgrp->kn);
586 }
587 
588 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
589 {
590 	pr_cont_kernfs_path(cgrp->kn);
591 }
592 
593 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
594 
595 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
596 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
597 int cgroup_rm_cftypes(struct cftype *cfts);
598 
599 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
600 
601 /*
602  * Control Group taskset, used to pass around set of tasks to cgroup_subsys
603  * methods.
604  */
605 struct cgroup_taskset;
606 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
607 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
608 
609 /**
610  * cgroup_taskset_for_each - iterate cgroup_taskset
611  * @task: the loop cursor
612  * @tset: taskset to iterate
613  */
614 #define cgroup_taskset_for_each(task, tset)				\
615 	for ((task) = cgroup_taskset_first((tset)); (task);		\
616 	     (task) = cgroup_taskset_next((tset)))
617 
618 /*
619  * Control Group subsystem type.
620  * See Documentation/cgroups/cgroups.txt for details
621  */
622 
623 struct cgroup_subsys {
624 	struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
625 	int (*css_online)(struct cgroup_subsys_state *css);
626 	void (*css_offline)(struct cgroup_subsys_state *css);
627 	void (*css_free)(struct cgroup_subsys_state *css);
628 	void (*css_reset)(struct cgroup_subsys_state *css);
629 
630 	int (*can_attach)(struct cgroup_subsys_state *css,
631 			  struct cgroup_taskset *tset);
632 	void (*cancel_attach)(struct cgroup_subsys_state *css,
633 			      struct cgroup_taskset *tset);
634 	void (*attach)(struct cgroup_subsys_state *css,
635 		       struct cgroup_taskset *tset);
636 	void (*fork)(struct task_struct *task);
637 	void (*exit)(struct cgroup_subsys_state *css,
638 		     struct cgroup_subsys_state *old_css,
639 		     struct task_struct *task);
640 	void (*bind)(struct cgroup_subsys_state *root_css);
641 
642 	int disabled;
643 	int early_init;
644 
645 	/*
646 	 * If %false, this subsystem is properly hierarchical -
647 	 * configuration, resource accounting and restriction on a parent
648 	 * cgroup cover those of its children.  If %true, hierarchy support
649 	 * is broken in some ways - some subsystems ignore hierarchy
650 	 * completely while others are only implemented half-way.
651 	 *
652 	 * It's now disallowed to create nested cgroups if the subsystem is
653 	 * broken and cgroup core will emit a warning message on such
654 	 * cases.  Eventually, all subsystems will be made properly
655 	 * hierarchical and this will go away.
656 	 */
657 	bool broken_hierarchy;
658 	bool warned_broken_hierarchy;
659 
660 	/* the following two fields are initialized automtically during boot */
661 	int id;
662 #define MAX_CGROUP_TYPE_NAMELEN 32
663 	const char *name;
664 
665 	/* link to parent, protected by cgroup_lock() */
666 	struct cgroup_root *root;
667 
668 	/* idr for css->id */
669 	struct idr css_idr;
670 
671 	/*
672 	 * List of cftypes.  Each entry is the first entry of an array
673 	 * terminated by zero length name.
674 	 */
675 	struct list_head cfts;
676 
677 	/*
678 	 * Base cftypes which are automatically registered.  The two can
679 	 * point to the same array.
680 	 */
681 	struct cftype *dfl_cftypes;	/* for the default hierarchy */
682 	struct cftype *legacy_cftypes;	/* for the legacy hierarchies */
683 
684 	/*
685 	 * A subsystem may depend on other subsystems.  When such subsystem
686 	 * is enabled on a cgroup, the depended-upon subsystems are enabled
687 	 * together if available.  Subsystems enabled due to dependency are
688 	 * not visible to userland until explicitly enabled.  The following
689 	 * specifies the mask of subsystems that this one depends on.
690 	 */
691 	unsigned int depends_on;
692 };
693 
694 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
695 #include <linux/cgroup_subsys.h>
696 #undef SUBSYS
697 
698 /**
699  * task_css_set_check - obtain a task's css_set with extra access conditions
700  * @task: the task to obtain css_set for
701  * @__c: extra condition expression to be passed to rcu_dereference_check()
702  *
703  * A task's css_set is RCU protected, initialized and exited while holding
704  * task_lock(), and can only be modified while holding both cgroup_mutex
705  * and task_lock() while the task is alive.  This macro verifies that the
706  * caller is inside proper critical section and returns @task's css_set.
707  *
708  * The caller can also specify additional allowed conditions via @__c, such
709  * as locks used during the cgroup_subsys::attach() methods.
710  */
711 #ifdef CONFIG_PROVE_RCU
712 extern struct mutex cgroup_mutex;
713 extern struct rw_semaphore css_set_rwsem;
714 #define task_css_set_check(task, __c)					\
715 	rcu_dereference_check((task)->cgroups,				\
716 		lockdep_is_held(&cgroup_mutex) ||			\
717 		lockdep_is_held(&css_set_rwsem) ||			\
718 		((task)->flags & PF_EXITING) || (__c))
719 #else
720 #define task_css_set_check(task, __c)					\
721 	rcu_dereference((task)->cgroups)
722 #endif
723 
724 /**
725  * task_css_check - obtain css for (task, subsys) w/ extra access conds
726  * @task: the target task
727  * @subsys_id: the target subsystem ID
728  * @__c: extra condition expression to be passed to rcu_dereference_check()
729  *
730  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
731  * synchronization rules are the same as task_css_set_check().
732  */
733 #define task_css_check(task, subsys_id, __c)				\
734 	task_css_set_check((task), (__c))->subsys[(subsys_id)]
735 
736 /**
737  * task_css_set - obtain a task's css_set
738  * @task: the task to obtain css_set for
739  *
740  * See task_css_set_check().
741  */
742 static inline struct css_set *task_css_set(struct task_struct *task)
743 {
744 	return task_css_set_check(task, false);
745 }
746 
747 /**
748  * task_css - obtain css for (task, subsys)
749  * @task: the target task
750  * @subsys_id: the target subsystem ID
751  *
752  * See task_css_check().
753  */
754 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
755 						   int subsys_id)
756 {
757 	return task_css_check(task, subsys_id, false);
758 }
759 
760 /**
761  * task_css_is_root - test whether a task belongs to the root css
762  * @task: the target task
763  * @subsys_id: the target subsystem ID
764  *
765  * Test whether @task belongs to the root css on the specified subsystem.
766  * May be invoked in any context.
767  */
768 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
769 {
770 	return task_css_check(task, subsys_id, true) ==
771 		init_css_set.subsys[subsys_id];
772 }
773 
774 static inline struct cgroup *task_cgroup(struct task_struct *task,
775 					 int subsys_id)
776 {
777 	return task_css(task, subsys_id)->cgroup;
778 }
779 
780 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
781 					   struct cgroup_subsys_state *parent);
782 
783 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
784 
785 /**
786  * css_for_each_child - iterate through children of a css
787  * @pos: the css * to use as the loop cursor
788  * @parent: css whose children to walk
789  *
790  * Walk @parent's children.  Must be called under rcu_read_lock().
791  *
792  * If a subsystem synchronizes ->css_online() and the start of iteration, a
793  * css which finished ->css_online() is guaranteed to be visible in the
794  * future iterations and will stay visible until the last reference is put.
795  * A css which hasn't finished ->css_online() or already finished
796  * ->css_offline() may show up during traversal.  It's each subsystem's
797  * responsibility to synchronize against on/offlining.
798  *
799  * It is allowed to temporarily drop RCU read lock during iteration.  The
800  * caller is responsible for ensuring that @pos remains accessible until
801  * the start of the next iteration by, for example, bumping the css refcnt.
802  */
803 #define css_for_each_child(pos, parent)					\
804 	for ((pos) = css_next_child(NULL, (parent)); (pos);		\
805 	     (pos) = css_next_child((pos), (parent)))
806 
807 struct cgroup_subsys_state *
808 css_next_descendant_pre(struct cgroup_subsys_state *pos,
809 			struct cgroup_subsys_state *css);
810 
811 struct cgroup_subsys_state *
812 css_rightmost_descendant(struct cgroup_subsys_state *pos);
813 
814 /**
815  * css_for_each_descendant_pre - pre-order walk of a css's descendants
816  * @pos: the css * to use as the loop cursor
817  * @root: css whose descendants to walk
818  *
819  * Walk @root's descendants.  @root is included in the iteration and the
820  * first node to be visited.  Must be called under rcu_read_lock().
821  *
822  * If a subsystem synchronizes ->css_online() and the start of iteration, a
823  * css which finished ->css_online() is guaranteed to be visible in the
824  * future iterations and will stay visible until the last reference is put.
825  * A css which hasn't finished ->css_online() or already finished
826  * ->css_offline() may show up during traversal.  It's each subsystem's
827  * responsibility to synchronize against on/offlining.
828  *
829  * For example, the following guarantees that a descendant can't escape
830  * state updates of its ancestors.
831  *
832  * my_online(@css)
833  * {
834  *	Lock @css's parent and @css;
835  *	Inherit state from the parent;
836  *	Unlock both.
837  * }
838  *
839  * my_update_state(@css)
840  * {
841  *	css_for_each_descendant_pre(@pos, @css) {
842  *		Lock @pos;
843  *		if (@pos == @css)
844  *			Update @css's state;
845  *		else
846  *			Verify @pos is alive and inherit state from its parent;
847  *		Unlock @pos;
848  *	}
849  * }
850  *
851  * As long as the inheriting step, including checking the parent state, is
852  * enclosed inside @pos locking, double-locking the parent isn't necessary
853  * while inheriting.  The state update to the parent is guaranteed to be
854  * visible by walking order and, as long as inheriting operations to the
855  * same @pos are atomic to each other, multiple updates racing each other
856  * still result in the correct state.  It's guaranateed that at least one
857  * inheritance happens for any css after the latest update to its parent.
858  *
859  * If checking parent's state requires locking the parent, each inheriting
860  * iteration should lock and unlock both @pos->parent and @pos.
861  *
862  * Alternatively, a subsystem may choose to use a single global lock to
863  * synchronize ->css_online() and ->css_offline() against tree-walking
864  * operations.
865  *
866  * It is allowed to temporarily drop RCU read lock during iteration.  The
867  * caller is responsible for ensuring that @pos remains accessible until
868  * the start of the next iteration by, for example, bumping the css refcnt.
869  */
870 #define css_for_each_descendant_pre(pos, css)				\
871 	for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);	\
872 	     (pos) = css_next_descendant_pre((pos), (css)))
873 
874 struct cgroup_subsys_state *
875 css_next_descendant_post(struct cgroup_subsys_state *pos,
876 			 struct cgroup_subsys_state *css);
877 
878 /**
879  * css_for_each_descendant_post - post-order walk of a css's descendants
880  * @pos: the css * to use as the loop cursor
881  * @css: css whose descendants to walk
882  *
883  * Similar to css_for_each_descendant_pre() but performs post-order
884  * traversal instead.  @root is included in the iteration and the last
885  * node to be visited.
886  *
887  * If a subsystem synchronizes ->css_online() and the start of iteration, a
888  * css which finished ->css_online() is guaranteed to be visible in the
889  * future iterations and will stay visible until the last reference is put.
890  * A css which hasn't finished ->css_online() or already finished
891  * ->css_offline() may show up during traversal.  It's each subsystem's
892  * responsibility to synchronize against on/offlining.
893  *
894  * Note that the walk visibility guarantee example described in pre-order
895  * walk doesn't apply the same to post-order walks.
896  */
897 #define css_for_each_descendant_post(pos, css)				\
898 	for ((pos) = css_next_descendant_post(NULL, (css)); (pos);	\
899 	     (pos) = css_next_descendant_post((pos), (css)))
900 
901 bool css_has_online_children(struct cgroup_subsys_state *css);
902 
903 /* A css_task_iter should be treated as an opaque object */
904 struct css_task_iter {
905 	struct cgroup_subsys		*ss;
906 
907 	struct list_head		*cset_pos;
908 	struct list_head		*cset_head;
909 
910 	struct list_head		*task_pos;
911 	struct list_head		*tasks_head;
912 	struct list_head		*mg_tasks_head;
913 };
914 
915 void css_task_iter_start(struct cgroup_subsys_state *css,
916 			 struct css_task_iter *it);
917 struct task_struct *css_task_iter_next(struct css_task_iter *it);
918 void css_task_iter_end(struct css_task_iter *it);
919 
920 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
921 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
922 
923 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
924 						       struct cgroup_subsys *ss);
925 
926 #else /* !CONFIG_CGROUPS */
927 
928 static inline int cgroup_init_early(void) { return 0; }
929 static inline int cgroup_init(void) { return 0; }
930 static inline void cgroup_fork(struct task_struct *p) {}
931 static inline void cgroup_post_fork(struct task_struct *p) {}
932 static inline void cgroup_exit(struct task_struct *p) {}
933 
934 static inline int cgroupstats_build(struct cgroupstats *stats,
935 					struct dentry *dentry)
936 {
937 	return -EINVAL;
938 }
939 
940 /* No cgroups - nothing to do */
941 static inline int cgroup_attach_task_all(struct task_struct *from,
942 					 struct task_struct *t)
943 {
944 	return 0;
945 }
946 
947 #endif /* !CONFIG_CGROUPS */
948 
949 #endif /* _LINUX_CGROUP_H */
950