xref: /openbmc/linux/include/linux/cgroup.h (revision bde24030b5a2825ef198b2f092b7271e22fae142)
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 #include <linux/fs.h>
23 #include <linux/percpu-refcount.h>
24 #include <linux/seq_file.h>
25 
26 #ifdef CONFIG_CGROUPS
27 
28 struct cgroupfs_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, int run_callbacks);
38 extern int cgroupstats_build(struct cgroupstats *stats,
39 				struct dentry *dentry);
40 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
41 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
42 
43 extern int proc_cgroup_show(struct seq_file *, void *);
44 
45 /*
46  * Define the enumeration of all cgroup subsystems.
47  *
48  * We define ids for builtin subsystems and then modular ones.
49  */
50 #define SUBSYS(_x) _x ## _subsys_id,
51 enum cgroup_subsys_id {
52 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
53 #include <linux/cgroup_subsys.h>
54 #undef IS_SUBSYS_ENABLED
55 	CGROUP_BUILTIN_SUBSYS_COUNT,
56 
57 	__CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
58 
59 #define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
60 #include <linux/cgroup_subsys.h>
61 #undef IS_SUBSYS_ENABLED
62 	CGROUP_SUBSYS_COUNT,
63 };
64 #undef SUBSYS
65 
66 /* Per-subsystem/per-cgroup state maintained by the system. */
67 struct cgroup_subsys_state {
68 	/* the cgroup that this css is attached to */
69 	struct cgroup *cgroup;
70 
71 	/* the cgroup subsystem that this css is attached to */
72 	struct cgroup_subsys *ss;
73 
74 	/* reference count - access via css_[try]get() and css_put() */
75 	struct percpu_ref refcnt;
76 
77 	/* the parent css */
78 	struct cgroup_subsys_state *parent;
79 
80 	unsigned long flags;
81 
82 	/* percpu_ref killing and RCU release */
83 	struct rcu_head rcu_head;
84 	struct work_struct destroy_work;
85 };
86 
87 /* bits in struct cgroup_subsys_state flags field */
88 enum {
89 	CSS_ROOT	= (1 << 0), /* this CSS is the root of the subsystem */
90 	CSS_ONLINE	= (1 << 1), /* between ->css_online() and ->css_offline() */
91 };
92 
93 /**
94  * css_get - obtain a reference on the specified css
95  * @css: target css
96  *
97  * The caller must already have a reference.
98  */
99 static inline void css_get(struct cgroup_subsys_state *css)
100 {
101 	/* We don't need to reference count the root state */
102 	if (!(css->flags & CSS_ROOT))
103 		percpu_ref_get(&css->refcnt);
104 }
105 
106 /**
107  * css_tryget - try to obtain a reference on the specified css
108  * @css: target css
109  *
110  * Obtain a reference on @css if it's alive.  The caller naturally needs to
111  * ensure that @css is accessible but doesn't have to be holding a
112  * reference on it - IOW, RCU protected access is good enough for this
113  * function.  Returns %true if a reference count was successfully obtained;
114  * %false otherwise.
115  */
116 static inline bool css_tryget(struct cgroup_subsys_state *css)
117 {
118 	if (css->flags & CSS_ROOT)
119 		return true;
120 	return percpu_ref_tryget(&css->refcnt);
121 }
122 
123 /**
124  * css_put - put a css reference
125  * @css: target css
126  *
127  * Put a reference obtained via css_get() and css_tryget().
128  */
129 static inline void css_put(struct cgroup_subsys_state *css)
130 {
131 	if (!(css->flags & CSS_ROOT))
132 		percpu_ref_put(&css->refcnt);
133 }
134 
135 /* bits in struct cgroup flags field */
136 enum {
137 	/* Control Group is dead */
138 	CGRP_DEAD,
139 	/*
140 	 * Control Group has previously had a child cgroup or a task,
141 	 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
142 	 */
143 	CGRP_RELEASABLE,
144 	/* Control Group requires release notifications to userspace */
145 	CGRP_NOTIFY_ON_RELEASE,
146 	/*
147 	 * Clone the parent's configuration when creating a new child
148 	 * cpuset cgroup.  For historical reasons, this option can be
149 	 * specified at mount time and thus is implemented here.
150 	 */
151 	CGRP_CPUSET_CLONE_CHILDREN,
152 	/* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
153 	CGRP_SANE_BEHAVIOR,
154 };
155 
156 struct cgroup_name {
157 	struct rcu_head rcu_head;
158 	char name[];
159 };
160 
161 struct cgroup {
162 	unsigned long flags;		/* "unsigned long" so bitops work */
163 
164 	/*
165 	 * idr allocated in-hierarchy ID.
166 	 *
167 	 * The ID of the root cgroup is always 0, and a new cgroup
168 	 * will be assigned with a smallest available ID.
169 	 */
170 	int id;
171 
172 	/* the number of attached css's */
173 	int nr_css;
174 
175 	/*
176 	 * We link our 'sibling' struct into our parent's 'children'.
177 	 * Our children link their 'sibling' into our 'children'.
178 	 */
179 	struct list_head sibling;	/* my parent's children */
180 	struct list_head children;	/* my children */
181 	struct list_head files;		/* my files */
182 
183 	struct cgroup *parent;		/* my parent */
184 	struct dentry *dentry;		/* cgroup fs entry, RCU protected */
185 
186 	/*
187 	 * Monotonically increasing unique serial number which defines a
188 	 * uniform order among all cgroups.  It's guaranteed that all
189 	 * ->children lists are in the ascending order of ->serial_nr.
190 	 * It's used to allow interrupting and resuming iterations.
191 	 */
192 	u64 serial_nr;
193 
194 	/*
195 	 * This is a copy of dentry->d_name, and it's needed because
196 	 * we can't use dentry->d_name in cgroup_path().
197 	 *
198 	 * You must acquire rcu_read_lock() to access cgrp->name, and
199 	 * the only place that can change it is rename(), which is
200 	 * protected by parent dir's i_mutex.
201 	 *
202 	 * Normally you should use cgroup_name() wrapper rather than
203 	 * access it directly.
204 	 */
205 	struct cgroup_name __rcu *name;
206 
207 	/* Private pointers for each registered subsystem */
208 	struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
209 
210 	struct cgroupfs_root *root;
211 
212 	/*
213 	 * List of cgrp_cset_links pointing at css_sets with tasks in this
214 	 * cgroup.  Protected by css_set_lock.
215 	 */
216 	struct list_head cset_links;
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 	/* dummy css with NULL ->ss, points back to this cgroup */
233 	struct cgroup_subsys_state dummy_css;
234 
235 	/* For css percpu_ref killing and RCU-protected deletion */
236 	struct rcu_head rcu_head;
237 	struct work_struct destroy_work;
238 
239 	/* directory xattrs */
240 	struct simple_xattrs xattrs;
241 };
242 
243 #define MAX_CGROUP_ROOT_NAMELEN 64
244 
245 /* cgroupfs_root->flags */
246 enum {
247 	/*
248 	 * Unfortunately, cgroup core and various controllers are riddled
249 	 * with idiosyncrasies and pointless options.  The following flag,
250 	 * when set, will force sane behavior - some options are forced on,
251 	 * others are disallowed, and some controllers will change their
252 	 * hierarchical or other behaviors.
253 	 *
254 	 * The set of behaviors affected by this flag are still being
255 	 * determined and developed and the mount option for this flag is
256 	 * prefixed with __DEVEL__.  The prefix will be dropped once we
257 	 * reach the point where all behaviors are compatible with the
258 	 * planned unified hierarchy, which will automatically turn on this
259 	 * flag.
260 	 *
261 	 * The followings are the behaviors currently affected this flag.
262 	 *
263 	 * - Mount options "noprefix" and "clone_children" are disallowed.
264 	 *   Also, cgroupfs file cgroup.clone_children is not created.
265 	 *
266 	 * - When mounting an existing superblock, mount options should
267 	 *   match.
268 	 *
269 	 * - Remount is disallowed.
270 	 *
271 	 * - rename(2) is disallowed.
272 	 *
273 	 * - "tasks" is removed.  Everything should be at process
274 	 *   granularity.  Use "cgroup.procs" instead.
275 	 *
276 	 * - "cgroup.procs" is not sorted.  pids will be unique unless they
277 	 *   got recycled inbetween reads.
278 	 *
279 	 * - "release_agent" and "notify_on_release" are removed.
280 	 *   Replacement notification mechanism will be implemented.
281 	 *
282 	 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
283 	 *   and take masks of ancestors with non-empty cpus/mems, instead of
284 	 *   being moved to an ancestor.
285 	 *
286 	 * - cpuset: a task can be moved into an empty cpuset, and again it
287 	 *   takes masks of ancestors.
288 	 *
289 	 * - memcg: use_hierarchy is on by default and the cgroup file for
290 	 *   the flag is not created.
291 	 *
292 	 * - blkcg: blk-throttle becomes properly hierarchical.
293 	 */
294 	CGRP_ROOT_SANE_BEHAVIOR	= (1 << 0),
295 
296 	CGRP_ROOT_NOPREFIX	= (1 << 1), /* mounted subsystems have no named prefix */
297 	CGRP_ROOT_XATTR		= (1 << 2), /* supports extended attributes */
298 
299 	/* mount options live below bit 16 */
300 	CGRP_ROOT_OPTION_MASK	= (1 << 16) - 1,
301 
302 	CGRP_ROOT_SUBSYS_BOUND	= (1 << 16), /* subsystems finished binding */
303 };
304 
305 /*
306  * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
307  * associated with a superblock to form an active hierarchy.  This is
308  * internal to cgroup core.  Don't access directly from controllers.
309  */
310 struct cgroupfs_root {
311 	struct super_block *sb;
312 
313 	/* The bitmask of subsystems attached to this hierarchy */
314 	unsigned long subsys_mask;
315 
316 	/* Unique id for this hierarchy. */
317 	int hierarchy_id;
318 
319 	/* The root cgroup for this hierarchy */
320 	struct cgroup top_cgroup;
321 
322 	/* Tracks how many cgroups are currently defined in hierarchy.*/
323 	int number_of_cgroups;
324 
325 	/* A list running through the active hierarchies */
326 	struct list_head root_list;
327 
328 	/* Hierarchy-specific flags */
329 	unsigned long flags;
330 
331 	/* IDs for cgroups in this hierarchy */
332 	struct idr cgroup_idr;
333 
334 	/* The path to use for release notifications. */
335 	char release_agent_path[PATH_MAX];
336 
337 	/* The name for this hierarchy - may be empty */
338 	char name[MAX_CGROUP_ROOT_NAMELEN];
339 };
340 
341 /*
342  * A css_set is a structure holding pointers to a set of
343  * cgroup_subsys_state objects. This saves space in the task struct
344  * object and speeds up fork()/exit(), since a single inc/dec and a
345  * list_add()/del() can bump the reference count on the entire cgroup
346  * set for a task.
347  */
348 
349 struct css_set {
350 
351 	/* Reference count */
352 	atomic_t refcount;
353 
354 	/*
355 	 * List running through all cgroup groups in the same hash
356 	 * slot. Protected by css_set_lock
357 	 */
358 	struct hlist_node hlist;
359 
360 	/*
361 	 * List running through all tasks using this cgroup
362 	 * group. Protected by css_set_lock
363 	 */
364 	struct list_head tasks;
365 
366 	/*
367 	 * List of cgrp_cset_links pointing at cgroups referenced from this
368 	 * css_set.  Protected by css_set_lock.
369 	 */
370 	struct list_head cgrp_links;
371 
372 	/*
373 	 * Set of subsystem states, one for each subsystem. This array
374 	 * is immutable after creation apart from the init_css_set
375 	 * during subsystem registration (at boot time) and modular subsystem
376 	 * loading/unloading.
377 	 */
378 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
379 
380 	/* For RCU-protected deletion */
381 	struct rcu_head rcu_head;
382 };
383 
384 /*
385  * struct cftype: handler definitions for cgroup control files
386  *
387  * When reading/writing to a file:
388  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
389  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
390  */
391 
392 /* cftype->flags */
393 enum {
394 	CFTYPE_ONLY_ON_ROOT	= (1 << 0),	/* only create on root cgrp */
395 	CFTYPE_NOT_ON_ROOT	= (1 << 1),	/* don't create on root cgrp */
396 	CFTYPE_INSANE		= (1 << 2),	/* don't create if sane_behavior */
397 	CFTYPE_NO_PREFIX	= (1 << 3),	/* (DON'T USE FOR NEW FILES) no subsys prefix */
398 };
399 
400 #define MAX_CFTYPE_NAME		64
401 
402 struct cftype {
403 	/*
404 	 * By convention, the name should begin with the name of the
405 	 * subsystem, followed by a period.  Zero length string indicates
406 	 * end of cftype array.
407 	 */
408 	char name[MAX_CFTYPE_NAME];
409 	int private;
410 	/*
411 	 * If not 0, file mode is set to this value, otherwise it will
412 	 * be figured out automatically
413 	 */
414 	umode_t mode;
415 
416 	/*
417 	 * If non-zero, defines the maximum length of string that can
418 	 * be passed to write_string; defaults to 64
419 	 */
420 	size_t max_write_len;
421 
422 	/* CFTYPE_* flags */
423 	unsigned int flags;
424 
425 	/*
426 	 * The subsys this file belongs to.  Initialized automatically
427 	 * during registration.  NULL for cgroup core files.
428 	 */
429 	struct cgroup_subsys *ss;
430 
431 	/*
432 	 * read_u64() is a shortcut for the common case of returning a
433 	 * single integer. Use it in place of read()
434 	 */
435 	u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
436 	/*
437 	 * read_s64() is a signed version of read_u64()
438 	 */
439 	s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
440 
441 	/* generic seq_file read interface */
442 	int (*seq_show)(struct seq_file *sf, void *v);
443 
444 	/* optional ops, implement all or none */
445 	void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
446 	void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
447 	void (*seq_stop)(struct seq_file *sf, void *v);
448 
449 	/*
450 	 * write_u64() is a shortcut for the common case of accepting
451 	 * a single integer (as parsed by simple_strtoull) from
452 	 * userspace. Use in place of write(); return 0 or error.
453 	 */
454 	int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
455 			 u64 val);
456 	/*
457 	 * write_s64() is a signed version of write_u64()
458 	 */
459 	int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
460 			 s64 val);
461 
462 	/*
463 	 * write_string() is passed a nul-terminated kernelspace
464 	 * buffer of maximum length determined by max_write_len.
465 	 * Returns 0 or -ve error code.
466 	 */
467 	int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft,
468 			    const char *buffer);
469 	/*
470 	 * trigger() callback can be used to get some kick from the
471 	 * userspace, when the actual string written is not important
472 	 * at all. The private field can be used to determine the
473 	 * kick type for multiplexing.
474 	 */
475 	int (*trigger)(struct cgroup_subsys_state *css, unsigned int event);
476 };
477 
478 /*
479  * cftype_sets describe cftypes belonging to a subsystem and are chained at
480  * cgroup_subsys->cftsets.  Each cftset points to an array of cftypes
481  * terminated by zero length name.
482  */
483 struct cftype_set {
484 	struct list_head		node;	/* chained at subsys->cftsets */
485 	struct cftype			*cfts;
486 };
487 
488 /*
489  * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.  Don't
490  * access directly.
491  */
492 struct cfent {
493 	struct list_head		node;
494 	struct dentry			*dentry;
495 	struct cftype			*type;
496 	struct cgroup_subsys_state	*css;
497 
498 	/* file xattrs */
499 	struct simple_xattrs		xattrs;
500 };
501 
502 /* seq_file->private points to the following, only ->priv is public */
503 struct cgroup_open_file {
504 	struct cfent			*cfe;
505 	void				*priv;
506 };
507 
508 /*
509  * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details.  This
510  * function can be called as long as @cgrp is accessible.
511  */
512 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
513 {
514 	return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
515 }
516 
517 /* Caller should hold rcu_read_lock() */
518 static inline const char *cgroup_name(const struct cgroup *cgrp)
519 {
520 	return rcu_dereference(cgrp->name)->name;
521 }
522 
523 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
524 {
525 	struct cgroup_open_file *of = seq->private;
526 	return of->cfe->css;
527 }
528 
529 static inline struct cftype *seq_cft(struct seq_file *seq)
530 {
531 	struct cgroup_open_file *of = seq->private;
532 	return of->cfe->type;
533 }
534 
535 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
536 int cgroup_rm_cftypes(struct cftype *cfts);
537 
538 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
539 
540 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
541 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
542 
543 int cgroup_task_count(const struct cgroup *cgrp);
544 
545 /*
546  * Control Group taskset, used to pass around set of tasks to cgroup_subsys
547  * methods.
548  */
549 struct cgroup_taskset;
550 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
551 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
552 struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset,
553 						   int subsys_id);
554 int cgroup_taskset_size(struct cgroup_taskset *tset);
555 
556 /**
557  * cgroup_taskset_for_each - iterate cgroup_taskset
558  * @task: the loop cursor
559  * @skip_css: skip if task's css matches this, %NULL to iterate through all
560  * @tset: taskset to iterate
561  */
562 #define cgroup_taskset_for_each(task, skip_css, tset)			\
563 	for ((task) = cgroup_taskset_first((tset)); (task);		\
564 	     (task) = cgroup_taskset_next((tset)))			\
565 		if (!(skip_css) ||					\
566 		    cgroup_taskset_cur_css((tset),			\
567 			(skip_css)->ss->subsys_id) != (skip_css))
568 
569 /*
570  * Control Group subsystem type.
571  * See Documentation/cgroups/cgroups.txt for details
572  */
573 
574 struct cgroup_subsys {
575 	struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
576 	int (*css_online)(struct cgroup_subsys_state *css);
577 	void (*css_offline)(struct cgroup_subsys_state *css);
578 	void (*css_free)(struct cgroup_subsys_state *css);
579 
580 	int (*can_attach)(struct cgroup_subsys_state *css,
581 			  struct cgroup_taskset *tset);
582 	void (*cancel_attach)(struct cgroup_subsys_state *css,
583 			      struct cgroup_taskset *tset);
584 	void (*attach)(struct cgroup_subsys_state *css,
585 		       struct cgroup_taskset *tset);
586 	void (*fork)(struct task_struct *task);
587 	void (*exit)(struct cgroup_subsys_state *css,
588 		     struct cgroup_subsys_state *old_css,
589 		     struct task_struct *task);
590 	void (*bind)(struct cgroup_subsys_state *root_css);
591 
592 	int subsys_id;
593 	int disabled;
594 	int early_init;
595 
596 	/*
597 	 * If %false, this subsystem is properly hierarchical -
598 	 * configuration, resource accounting and restriction on a parent
599 	 * cgroup cover those of its children.  If %true, hierarchy support
600 	 * is broken in some ways - some subsystems ignore hierarchy
601 	 * completely while others are only implemented half-way.
602 	 *
603 	 * It's now disallowed to create nested cgroups if the subsystem is
604 	 * broken and cgroup core will emit a warning message on such
605 	 * cases.  Eventually, all subsystems will be made properly
606 	 * hierarchical and this will go away.
607 	 */
608 	bool broken_hierarchy;
609 	bool warned_broken_hierarchy;
610 
611 #define MAX_CGROUP_TYPE_NAMELEN 32
612 	const char *name;
613 
614 	/* link to parent, protected by cgroup_lock() */
615 	struct cgroupfs_root *root;
616 
617 	/* list of cftype_sets */
618 	struct list_head cftsets;
619 
620 	/* base cftypes, automatically [de]registered with subsys itself */
621 	struct cftype *base_cftypes;
622 	struct cftype_set base_cftset;
623 
624 	/* should be defined only by modular subsystems */
625 	struct module *module;
626 };
627 
628 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
629 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
630 #include <linux/cgroup_subsys.h>
631 #undef IS_SUBSYS_ENABLED
632 #undef SUBSYS
633 
634 /**
635  * css_parent - find the parent css
636  * @css: the target cgroup_subsys_state
637  *
638  * Return the parent css of @css.  This function is guaranteed to return
639  * non-NULL parent as long as @css isn't the root.
640  */
641 static inline
642 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
643 {
644 	return css->parent;
645 }
646 
647 /**
648  * task_css_set_check - obtain a task's css_set with extra access conditions
649  * @task: the task to obtain css_set for
650  * @__c: extra condition expression to be passed to rcu_dereference_check()
651  *
652  * A task's css_set is RCU protected, initialized and exited while holding
653  * task_lock(), and can only be modified while holding both cgroup_mutex
654  * and task_lock() while the task is alive.  This macro verifies that the
655  * caller is inside proper critical section and returns @task's css_set.
656  *
657  * The caller can also specify additional allowed conditions via @__c, such
658  * as locks used during the cgroup_subsys::attach() methods.
659  */
660 #ifdef CONFIG_PROVE_RCU
661 extern struct mutex cgroup_mutex;
662 #define task_css_set_check(task, __c)					\
663 	rcu_dereference_check((task)->cgroups,				\
664 		lockdep_is_held(&(task)->alloc_lock) ||			\
665 		lockdep_is_held(&cgroup_mutex) || (__c))
666 #else
667 #define task_css_set_check(task, __c)					\
668 	rcu_dereference((task)->cgroups)
669 #endif
670 
671 /**
672  * task_css_check - obtain css for (task, subsys) w/ extra access conds
673  * @task: the target task
674  * @subsys_id: the target subsystem ID
675  * @__c: extra condition expression to be passed to rcu_dereference_check()
676  *
677  * Return the cgroup_subsys_state for the (@task, @subsys_id) pair.  The
678  * synchronization rules are the same as task_css_set_check().
679  */
680 #define task_css_check(task, subsys_id, __c)				\
681 	task_css_set_check((task), (__c))->subsys[(subsys_id)]
682 
683 /**
684  * task_css_set - obtain a task's css_set
685  * @task: the task to obtain css_set for
686  *
687  * See task_css_set_check().
688  */
689 static inline struct css_set *task_css_set(struct task_struct *task)
690 {
691 	return task_css_set_check(task, false);
692 }
693 
694 /**
695  * task_css - obtain css for (task, subsys)
696  * @task: the target task
697  * @subsys_id: the target subsystem ID
698  *
699  * See task_css_check().
700  */
701 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
702 						   int subsys_id)
703 {
704 	return task_css_check(task, subsys_id, false);
705 }
706 
707 static inline struct cgroup *task_cgroup(struct task_struct *task,
708 					 int subsys_id)
709 {
710 	return task_css(task, subsys_id)->cgroup;
711 }
712 
713 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
714 					   struct cgroup_subsys_state *parent);
715 
716 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
717 
718 /**
719  * css_for_each_child - iterate through children of a css
720  * @pos: the css * to use as the loop cursor
721  * @parent: css whose children to walk
722  *
723  * Walk @parent's children.  Must be called under rcu_read_lock().  A child
724  * css which hasn't finished ->css_online() or already has finished
725  * ->css_offline() may show up during traversal and it's each subsystem's
726  * responsibility to verify that each @pos is alive.
727  *
728  * If a subsystem synchronizes against the parent in its ->css_online() and
729  * before starting iterating, a css which finished ->css_online() is
730  * guaranteed to be visible in the future iterations.
731  *
732  * It is allowed to temporarily drop RCU read lock during iteration.  The
733  * caller is responsible for ensuring that @pos remains accessible until
734  * the start of the next iteration by, for example, bumping the css refcnt.
735  */
736 #define css_for_each_child(pos, parent)					\
737 	for ((pos) = css_next_child(NULL, (parent)); (pos);		\
738 	     (pos) = css_next_child((pos), (parent)))
739 
740 struct cgroup_subsys_state *
741 css_next_descendant_pre(struct cgroup_subsys_state *pos,
742 			struct cgroup_subsys_state *css);
743 
744 struct cgroup_subsys_state *
745 css_rightmost_descendant(struct cgroup_subsys_state *pos);
746 
747 /**
748  * css_for_each_descendant_pre - pre-order walk of a css's descendants
749  * @pos: the css * to use as the loop cursor
750  * @root: css whose descendants to walk
751  *
752  * Walk @root's descendants.  @root is included in the iteration and the
753  * first node to be visited.  Must be called under rcu_read_lock().  A
754  * descendant css which hasn't finished ->css_online() or already has
755  * finished ->css_offline() may show up during traversal and it's each
756  * subsystem's responsibility to verify that each @pos is alive.
757  *
758  * If a subsystem synchronizes against the parent in its ->css_online() and
759  * before starting iterating, and synchronizes against @pos on each
760  * iteration, any descendant css which finished ->css_online() is
761  * guaranteed to be visible in the future iterations.
762  *
763  * In other words, the following guarantees that a descendant can't escape
764  * state updates of its ancestors.
765  *
766  * my_online(@css)
767  * {
768  *	Lock @css's parent and @css;
769  *	Inherit state from the parent;
770  *	Unlock both.
771  * }
772  *
773  * my_update_state(@css)
774  * {
775  *	css_for_each_descendant_pre(@pos, @css) {
776  *		Lock @pos;
777  *		if (@pos == @css)
778  *			Update @css's state;
779  *		else
780  *			Verify @pos is alive and inherit state from its parent;
781  *		Unlock @pos;
782  *	}
783  * }
784  *
785  * As long as the inheriting step, including checking the parent state, is
786  * enclosed inside @pos locking, double-locking the parent isn't necessary
787  * while inheriting.  The state update to the parent is guaranteed to be
788  * visible by walking order and, as long as inheriting operations to the
789  * same @pos are atomic to each other, multiple updates racing each other
790  * still result in the correct state.  It's guaranateed that at least one
791  * inheritance happens for any css after the latest update to its parent.
792  *
793  * If checking parent's state requires locking the parent, each inheriting
794  * iteration should lock and unlock both @pos->parent and @pos.
795  *
796  * Alternatively, a subsystem may choose to use a single global lock to
797  * synchronize ->css_online() and ->css_offline() against tree-walking
798  * operations.
799  *
800  * It is allowed to temporarily drop RCU read lock during iteration.  The
801  * caller is responsible for ensuring that @pos remains accessible until
802  * the start of the next iteration by, for example, bumping the css refcnt.
803  */
804 #define css_for_each_descendant_pre(pos, css)				\
805 	for ((pos) = css_next_descendant_pre(NULL, (css)); (pos);	\
806 	     (pos) = css_next_descendant_pre((pos), (css)))
807 
808 struct cgroup_subsys_state *
809 css_next_descendant_post(struct cgroup_subsys_state *pos,
810 			 struct cgroup_subsys_state *css);
811 
812 /**
813  * css_for_each_descendant_post - post-order walk of a css's descendants
814  * @pos: the css * to use as the loop cursor
815  * @css: css whose descendants to walk
816  *
817  * Similar to css_for_each_descendant_pre() but performs post-order
818  * traversal instead.  @root is included in the iteration and the last
819  * node to be visited.  Note that the walk visibility guarantee described
820  * in pre-order walk doesn't apply the same to post-order walks.
821  */
822 #define css_for_each_descendant_post(pos, css)				\
823 	for ((pos) = css_next_descendant_post(NULL, (css)); (pos);	\
824 	     (pos) = css_next_descendant_post((pos), (css)))
825 
826 /* A css_task_iter should be treated as an opaque object */
827 struct css_task_iter {
828 	struct cgroup_subsys_state	*origin_css;
829 	struct list_head		*cset_link;
830 	struct list_head		*task;
831 };
832 
833 void css_task_iter_start(struct cgroup_subsys_state *css,
834 			 struct css_task_iter *it);
835 struct task_struct *css_task_iter_next(struct css_task_iter *it);
836 void css_task_iter_end(struct css_task_iter *it);
837 
838 int css_scan_tasks(struct cgroup_subsys_state *css,
839 		   bool (*test)(struct task_struct *, void *),
840 		   void (*process)(struct task_struct *, void *),
841 		   void *data, struct ptr_heap *heap);
842 
843 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
844 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
845 
846 struct cgroup_subsys_state *css_from_dir(struct dentry *dentry,
847 					 struct cgroup_subsys *ss);
848 
849 #else /* !CONFIG_CGROUPS */
850 
851 static inline int cgroup_init_early(void) { return 0; }
852 static inline int cgroup_init(void) { return 0; }
853 static inline void cgroup_fork(struct task_struct *p) {}
854 static inline void cgroup_post_fork(struct task_struct *p) {}
855 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
856 
857 static inline int cgroupstats_build(struct cgroupstats *stats,
858 					struct dentry *dentry)
859 {
860 	return -EINVAL;
861 }
862 
863 /* No cgroups - nothing to do */
864 static inline int cgroup_attach_task_all(struct task_struct *from,
865 					 struct task_struct *t)
866 {
867 	return 0;
868 }
869 
870 #endif /* !CONFIG_CGROUPS */
871 
872 #endif /* _LINUX_CGROUP_H */
873