1 #ifndef _LINUX_CGROUP_H
2 #define _LINUX_CGROUP_H
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
11 #include <linux/sched.h>
12 #include <linux/cpumask.h>
13 #include <linux/nodemask.h>
14 #include <linux/rcupdate.h>
15 #include <linux/rculist.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/rwsem.h>
18 #include <linux/idr.h>
19 #include <linux/workqueue.h>
21 #include <linux/percpu-refcount.h>
22 #include <linux/seq_file.h>
23 #include <linux/kernfs.h>
24 #include <linux/wait.h>
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);
41 extern int proc_cgroup_show(struct seq_file *, void *);
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>
52 * Per-subsystem/per-cgroup state maintained by the system. This is the
53 * fundamental structural building block that controllers deal with.
55 * Fields marked with "PI:" are public and immutable and may be accessed
56 * directly without synchronization.
58 struct cgroup_subsys_state {
59 /* PI: the cgroup that this css is attached to */
60 struct cgroup *cgroup;
62 /* PI: the cgroup subsystem that this css is attached to */
63 struct cgroup_subsys *ss;
65 /* reference count - access via css_[try]get() and css_put() */
66 struct percpu_ref refcnt;
68 /* PI: the parent css */
69 struct cgroup_subsys_state *parent;
71 /* siblings list anchored at the parent's ->children */
72 struct list_head sibling;
73 struct list_head children;
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().
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.
91 /* percpu_ref killing and RCU release */
92 struct rcu_head rcu_head;
93 struct work_struct destroy_work;
96 /* bits in struct cgroup_subsys_state flags field */
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 */
104 * css_get - obtain a reference on the specified css
107 * The caller must already have a reference.
109 static inline void css_get(struct cgroup_subsys_state *css)
111 if (!(css->flags & CSS_NO_REF))
112 percpu_ref_get(&css->refcnt);
116 * css_tryget - try to obtain a reference on the specified css
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.
126 static inline bool css_tryget(struct cgroup_subsys_state *css)
128 if (!(css->flags & CSS_NO_REF))
129 return percpu_ref_tryget(&css->refcnt);
134 * css_tryget_online - try to obtain a reference on the specified css if online
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;
143 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
145 if (!(css->flags & CSS_NO_REF))
146 return percpu_ref_tryget_live(&css->refcnt);
151 * css_put - put a css reference
154 * Put a reference obtained via css_get() and css_tryget_online().
156 static inline void css_put(struct cgroup_subsys_state *css)
158 if (!(css->flags & CSS_NO_REF))
159 percpu_ref_put(&css->refcnt);
162 /* bits in struct cgroup flags field */
165 * Control Group has previously had a child cgroup or a task,
166 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
169 /* Control Group requires release notifications to userspace */
170 CGRP_NOTIFY_ON_RELEASE,
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.
176 CGRP_CPUSET_CLONE_CHILDREN,
180 /* self css with NULL ->ss, points back to this cgroup */
181 struct cgroup_subsys_state self;
183 unsigned long flags; /* "unsigned long" so bitops work */
186 * idr allocated in-hierarchy ID.
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.
191 * Allocating/Removing ID must be protected by cgroup_mutex.
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.
203 struct kernfs_node *kn; /* cgroup kernfs entry */
204 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
206 /* the bitmask of subsystems enabled on the child cgroups */
207 unsigned int child_subsys_mask;
209 /* Private pointers for each registered subsystem */
210 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
212 struct cgroup_root *root;
215 * List of cgrp_cset_links pointing at css_sets with tasks in this
216 * cgroup. Protected by css_set_lock.
218 struct list_head cset_links;
221 * On the default hierarchy, a css_set for a cgroup with some
222 * susbsys disabled will point to css's which are associated with
223 * the closest ancestor which has the subsys enabled. The
224 * following lists all css_sets which point to this cgroup's css
225 * for the given subsystem.
227 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
230 * Linked list running through all cgroups that can
231 * potentially be reaped by the release agent. Protected by
234 struct list_head release_list;
237 * list of pidlists, up to two for each namespace (one for procs, one
238 * for tasks); created on demand.
240 struct list_head pidlists;
241 struct mutex pidlist_mutex;
243 /* used to wait for offlining of csses */
244 wait_queue_head_t offline_waitq;
247 #define MAX_CGROUP_ROOT_NAMELEN 64
249 /* cgroup_root->flags */
252 * Unfortunately, cgroup core and various controllers are riddled
253 * with idiosyncrasies and pointless options. The following flag,
254 * when set, will force sane behavior - some options are forced on,
255 * others are disallowed, and some controllers will change their
256 * hierarchical or other behaviors.
258 * The set of behaviors affected by this flag are still being
259 * determined and developed and the mount option for this flag is
260 * prefixed with __DEVEL__. The prefix will be dropped once we
261 * reach the point where all behaviors are compatible with the
262 * planned unified hierarchy, which will automatically turn on this
265 * The followings are the behaviors currently affected this flag.
267 * - Mount options "noprefix", "xattr", "clone_children",
268 * "release_agent" and "name" are disallowed.
270 * - When mounting an existing superblock, mount options should
273 * - Remount is disallowed.
275 * - rename(2) is disallowed.
277 * - "tasks" is removed. Everything should be at process
278 * granularity. Use "cgroup.procs" instead.
280 * - "cgroup.procs" is not sorted. pids will be unique unless they
281 * got recycled inbetween reads.
283 * - "release_agent" and "notify_on_release" are removed.
284 * Replacement notification mechanism will be implemented.
286 * - "cgroup.clone_children" is removed.
288 * - "cgroup.subtree_populated" is available. Its value is 0 if
289 * the cgroup and its descendants contain no task; otherwise, 1.
290 * The file also generates kernfs notification which can be
291 * monitored through poll and [di]notify when the value of the
294 * - If mount is requested with sane_behavior but without any
295 * subsystem, the default unified hierarchy is mounted.
297 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
298 * and take masks of ancestors with non-empty cpus/mems, instead of
299 * being moved to an ancestor.
301 * - cpuset: a task can be moved into an empty cpuset, and again it
302 * takes masks of ancestors.
304 * - memcg: use_hierarchy is on by default and the cgroup file for
305 * the flag is not created.
307 * - blkcg: blk-throttle becomes properly hierarchical.
309 * - debug: disallowed on the default hierarchy.
311 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
313 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
314 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
316 /* mount options live below bit 16 */
317 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
321 * A cgroup_root represents the root of a cgroup hierarchy, and may be
322 * associated with a kernfs_root to form an active hierarchy. This is
323 * internal to cgroup core. Don't access directly from controllers.
326 struct kernfs_root *kf_root;
328 /* The bitmask of subsystems attached to this hierarchy */
329 unsigned int subsys_mask;
331 /* Unique id for this hierarchy. */
334 /* The root cgroup. Root is destroyed on its release. */
337 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
340 /* A list running through the active hierarchies */
341 struct list_head root_list;
343 /* Hierarchy-specific flags */
346 /* IDs for cgroups in this hierarchy */
347 struct idr cgroup_idr;
349 /* The path to use for release notifications. */
350 char release_agent_path[PATH_MAX];
352 /* The name for this hierarchy - may be empty */
353 char name[MAX_CGROUP_ROOT_NAMELEN];
357 * A css_set is a structure holding pointers to a set of
358 * cgroup_subsys_state objects. This saves space in the task struct
359 * object and speeds up fork()/exit(), since a single inc/dec and a
360 * list_add()/del() can bump the reference count on the entire cgroup
366 /* Reference count */
370 * List running through all cgroup groups in the same hash
371 * slot. Protected by css_set_lock
373 struct hlist_node hlist;
376 * Lists running through all tasks using this cgroup group.
377 * mg_tasks lists tasks which belong to this cset but are in the
378 * process of being migrated out or in. Protected by
379 * css_set_rwsem, but, during migration, once tasks are moved to
380 * mg_tasks, it can be read safely while holding cgroup_mutex.
382 struct list_head tasks;
383 struct list_head mg_tasks;
386 * List of cgrp_cset_links pointing at cgroups referenced from this
387 * css_set. Protected by css_set_lock.
389 struct list_head cgrp_links;
391 /* the default cgroup associated with this css_set */
392 struct cgroup *dfl_cgrp;
395 * Set of subsystem states, one for each subsystem. This array is
396 * immutable after creation apart from the init_css_set during
397 * subsystem registration (at boot time).
399 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
402 * List of csets participating in the on-going migration either as
403 * source or destination. Protected by cgroup_mutex.
405 struct list_head mg_preload_node;
406 struct list_head mg_node;
409 * If this cset is acting as the source of migration the following
410 * two fields are set. mg_src_cgrp is the source cgroup of the
411 * on-going migration and mg_dst_cset is the destination cset the
412 * target tasks on this cset should be migrated to. Protected by
415 struct cgroup *mg_src_cgrp;
416 struct css_set *mg_dst_cset;
419 * On the default hierarhcy, ->subsys[ssid] may point to a css
420 * attached to an ancestor instead of the cgroup this css_set is
421 * associated with. The following node is anchored at
422 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
423 * iterate through all css's attached to a given cgroup.
425 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
427 /* For RCU-protected deletion */
428 struct rcu_head rcu_head;
432 * struct cftype: handler definitions for cgroup control files
434 * When reading/writing to a file:
435 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
436 * - the 'cftype' of the file is file->f_dentry->d_fsdata
441 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
442 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
443 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
444 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
445 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
448 #define MAX_CFTYPE_NAME 64
452 * By convention, the name should begin with the name of the
453 * subsystem, followed by a period. Zero length string indicates
454 * end of cftype array.
456 char name[MAX_CFTYPE_NAME];
459 * If not 0, file mode is set to this value, otherwise it will
460 * be figured out automatically
465 * The maximum length of string, excluding trailing nul, that can
466 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
468 size_t max_write_len;
474 * Fields used for internal bookkeeping. Initialized automatically
475 * during registration.
477 struct cgroup_subsys *ss; /* NULL for cgroup core files */
478 struct list_head node; /* anchored at ss->cfts */
479 struct kernfs_ops *kf_ops;
482 * read_u64() is a shortcut for the common case of returning a
483 * single integer. Use it in place of read()
485 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
487 * read_s64() is a signed version of read_u64()
489 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
491 /* generic seq_file read interface */
492 int (*seq_show)(struct seq_file *sf, void *v);
494 /* optional ops, implement all or none */
495 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
496 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
497 void (*seq_stop)(struct seq_file *sf, void *v);
500 * write_u64() is a shortcut for the common case of accepting
501 * a single integer (as parsed by simple_strtoull) from
502 * userspace. Use in place of write(); return 0 or error.
504 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
507 * write_s64() is a signed version of write_u64()
509 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
513 * write() is the generic write callback which maps directly to
514 * kernfs write operation and overrides all other operations.
515 * Maximum write size is determined by ->max_write_len. Use
516 * of_css/cft() to access the associated css and cft.
518 ssize_t (*write)(struct kernfs_open_file *of,
519 char *buf, size_t nbytes, loff_t off);
521 #ifdef CONFIG_DEBUG_LOCK_ALLOC
522 struct lock_class_key lockdep_key;
526 extern struct cgroup_root cgrp_dfl_root;
527 extern struct css_set init_css_set;
529 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
531 return cgrp->root == &cgrp_dfl_root;
535 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
536 * function can be called as long as @cgrp is accessible.
538 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
540 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
543 /* no synchronization, the result can only be used as a hint */
544 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
546 return !list_empty(&cgrp->cset_links);
549 /* returns ino associated with a cgroup, 0 indicates unmounted root */
550 static inline ino_t cgroup_ino(struct cgroup *cgrp)
553 return cgrp->kn->ino;
558 /* cft/css accessors for cftype->write() operation */
559 static inline struct cftype *of_cft(struct kernfs_open_file *of)
564 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
566 /* cft/css accessors for cftype->seq_*() operations */
567 static inline struct cftype *seq_cft(struct seq_file *seq)
569 return of_cft(seq->private);
572 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
574 return of_css(seq->private);
578 * Name / path handling functions. All are thin wrappers around the kernfs
579 * counterparts and can be called under any context.
582 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
584 return kernfs_name(cgrp->kn, buf, buflen);
587 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
590 return kernfs_path(cgrp->kn, buf, buflen);
593 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
595 pr_cont_kernfs_name(cgrp->kn);
598 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
600 pr_cont_kernfs_path(cgrp->kn);
603 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
605 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
606 int cgroup_rm_cftypes(struct cftype *cfts);
608 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
611 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
614 struct cgroup_taskset;
615 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
616 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
619 * cgroup_taskset_for_each - iterate cgroup_taskset
620 * @task: the loop cursor
621 * @tset: taskset to iterate
623 #define cgroup_taskset_for_each(task, tset) \
624 for ((task) = cgroup_taskset_first((tset)); (task); \
625 (task) = cgroup_taskset_next((tset)))
628 * Control Group subsystem type.
629 * See Documentation/cgroups/cgroups.txt for details
632 struct cgroup_subsys {
633 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
634 int (*css_online)(struct cgroup_subsys_state *css);
635 void (*css_offline)(struct cgroup_subsys_state *css);
636 void (*css_free)(struct cgroup_subsys_state *css);
638 int (*can_attach)(struct cgroup_subsys_state *css,
639 struct cgroup_taskset *tset);
640 void (*cancel_attach)(struct cgroup_subsys_state *css,
641 struct cgroup_taskset *tset);
642 void (*attach)(struct cgroup_subsys_state *css,
643 struct cgroup_taskset *tset);
644 void (*fork)(struct task_struct *task);
645 void (*exit)(struct cgroup_subsys_state *css,
646 struct cgroup_subsys_state *old_css,
647 struct task_struct *task);
648 void (*bind)(struct cgroup_subsys_state *root_css);
654 * If %false, this subsystem is properly hierarchical -
655 * configuration, resource accounting and restriction on a parent
656 * cgroup cover those of its children. If %true, hierarchy support
657 * is broken in some ways - some subsystems ignore hierarchy
658 * completely while others are only implemented half-way.
660 * It's now disallowed to create nested cgroups if the subsystem is
661 * broken and cgroup core will emit a warning message on such
662 * cases. Eventually, all subsystems will be made properly
663 * hierarchical and this will go away.
665 bool broken_hierarchy;
666 bool warned_broken_hierarchy;
668 /* the following two fields are initialized automtically during boot */
670 #define MAX_CGROUP_TYPE_NAMELEN 32
673 /* link to parent, protected by cgroup_lock() */
674 struct cgroup_root *root;
676 /* idr for css->id */
680 * List of cftypes. Each entry is the first entry of an array
681 * terminated by zero length name.
683 struct list_head cfts;
685 /* base cftypes, automatically registered with subsys itself */
686 struct cftype *base_cftypes;
689 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
690 #include <linux/cgroup_subsys.h>
694 * task_css_set_check - obtain a task's css_set with extra access conditions
695 * @task: the task to obtain css_set for
696 * @__c: extra condition expression to be passed to rcu_dereference_check()
698 * A task's css_set is RCU protected, initialized and exited while holding
699 * task_lock(), and can only be modified while holding both cgroup_mutex
700 * and task_lock() while the task is alive. This macro verifies that the
701 * caller is inside proper critical section and returns @task's css_set.
703 * The caller can also specify additional allowed conditions via @__c, such
704 * as locks used during the cgroup_subsys::attach() methods.
706 #ifdef CONFIG_PROVE_RCU
707 extern struct mutex cgroup_mutex;
708 extern struct rw_semaphore css_set_rwsem;
709 #define task_css_set_check(task, __c) \
710 rcu_dereference_check((task)->cgroups, \
711 lockdep_is_held(&cgroup_mutex) || \
712 lockdep_is_held(&css_set_rwsem) || \
713 ((task)->flags & PF_EXITING) || (__c))
715 #define task_css_set_check(task, __c) \
716 rcu_dereference((task)->cgroups)
720 * task_css_check - obtain css for (task, subsys) w/ extra access conds
721 * @task: the target task
722 * @subsys_id: the target subsystem ID
723 * @__c: extra condition expression to be passed to rcu_dereference_check()
725 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
726 * synchronization rules are the same as task_css_set_check().
728 #define task_css_check(task, subsys_id, __c) \
729 task_css_set_check((task), (__c))->subsys[(subsys_id)]
732 * task_css_set - obtain a task's css_set
733 * @task: the task to obtain css_set for
735 * See task_css_set_check().
737 static inline struct css_set *task_css_set(struct task_struct *task)
739 return task_css_set_check(task, false);
743 * task_css - obtain css for (task, subsys)
744 * @task: the target task
745 * @subsys_id: the target subsystem ID
747 * See task_css_check().
749 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
752 return task_css_check(task, subsys_id, false);
756 * task_css_is_root - test whether a task belongs to the root css
757 * @task: the target task
758 * @subsys_id: the target subsystem ID
760 * Test whether @task belongs to the root css on the specified subsystem.
761 * May be invoked in any context.
763 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
765 return task_css_check(task, subsys_id, true) ==
766 init_css_set.subsys[subsys_id];
769 static inline struct cgroup *task_cgroup(struct task_struct *task,
772 return task_css(task, subsys_id)->cgroup;
775 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
776 struct cgroup_subsys_state *parent);
778 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
781 * css_for_each_child - iterate through children of a css
782 * @pos: the css * to use as the loop cursor
783 * @parent: css whose children to walk
785 * Walk @parent's children. Must be called under rcu_read_lock().
787 * If a subsystem synchronizes ->css_online() and the start of iteration, a
788 * css which finished ->css_online() is guaranteed to be visible in the
789 * future iterations and will stay visible until the last reference is put.
790 * A css which hasn't finished ->css_online() or already finished
791 * ->css_offline() may show up during traversal. It's each subsystem's
792 * responsibility to synchronize against on/offlining.
794 * It is allowed to temporarily drop RCU read lock during iteration. The
795 * caller is responsible for ensuring that @pos remains accessible until
796 * the start of the next iteration by, for example, bumping the css refcnt.
798 #define css_for_each_child(pos, parent) \
799 for ((pos) = css_next_child(NULL, (parent)); (pos); \
800 (pos) = css_next_child((pos), (parent)))
802 struct cgroup_subsys_state *
803 css_next_descendant_pre(struct cgroup_subsys_state *pos,
804 struct cgroup_subsys_state *css);
806 struct cgroup_subsys_state *
807 css_rightmost_descendant(struct cgroup_subsys_state *pos);
810 * css_for_each_descendant_pre - pre-order walk of a css's descendants
811 * @pos: the css * to use as the loop cursor
812 * @root: css whose descendants to walk
814 * Walk @root's descendants. @root is included in the iteration and the
815 * first node to be visited. Must be called under rcu_read_lock().
817 * If a subsystem synchronizes ->css_online() and the start of iteration, a
818 * css which finished ->css_online() is guaranteed to be visible in the
819 * future iterations and will stay visible until the last reference is put.
820 * A css which hasn't finished ->css_online() or already finished
821 * ->css_offline() may show up during traversal. It's each subsystem's
822 * responsibility to synchronize against on/offlining.
824 * For example, the following guarantees that a descendant can't escape
825 * state updates of its ancestors.
829 * Lock @css's parent and @css;
830 * Inherit state from the parent;
834 * my_update_state(@css)
836 * css_for_each_descendant_pre(@pos, @css) {
839 * Update @css's state;
841 * Verify @pos is alive and inherit state from its parent;
846 * As long as the inheriting step, including checking the parent state, is
847 * enclosed inside @pos locking, double-locking the parent isn't necessary
848 * while inheriting. The state update to the parent is guaranteed to be
849 * visible by walking order and, as long as inheriting operations to the
850 * same @pos are atomic to each other, multiple updates racing each other
851 * still result in the correct state. It's guaranateed that at least one
852 * inheritance happens for any css after the latest update to its parent.
854 * If checking parent's state requires locking the parent, each inheriting
855 * iteration should lock and unlock both @pos->parent and @pos.
857 * Alternatively, a subsystem may choose to use a single global lock to
858 * synchronize ->css_online() and ->css_offline() against tree-walking
861 * It is allowed to temporarily drop RCU read lock during iteration. The
862 * caller is responsible for ensuring that @pos remains accessible until
863 * the start of the next iteration by, for example, bumping the css refcnt.
865 #define css_for_each_descendant_pre(pos, css) \
866 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
867 (pos) = css_next_descendant_pre((pos), (css)))
869 struct cgroup_subsys_state *
870 css_next_descendant_post(struct cgroup_subsys_state *pos,
871 struct cgroup_subsys_state *css);
874 * css_for_each_descendant_post - post-order walk of a css's descendants
875 * @pos: the css * to use as the loop cursor
876 * @css: css whose descendants to walk
878 * Similar to css_for_each_descendant_pre() but performs post-order
879 * traversal instead. @root is included in the iteration and the last
880 * node to be visited.
882 * If a subsystem synchronizes ->css_online() and the start of iteration, a
883 * css which finished ->css_online() is guaranteed to be visible in the
884 * future iterations and will stay visible until the last reference is put.
885 * A css which hasn't finished ->css_online() or already finished
886 * ->css_offline() may show up during traversal. It's each subsystem's
887 * responsibility to synchronize against on/offlining.
889 * Note that the walk visibility guarantee example described in pre-order
890 * walk doesn't apply the same to post-order walks.
892 #define css_for_each_descendant_post(pos, css) \
893 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
894 (pos) = css_next_descendant_post((pos), (css)))
896 bool css_has_online_children(struct cgroup_subsys_state *css);
898 /* A css_task_iter should be treated as an opaque object */
899 struct css_task_iter {
900 struct cgroup_subsys *ss;
902 struct list_head *cset_pos;
903 struct list_head *cset_head;
905 struct list_head *task_pos;
906 struct list_head *tasks_head;
907 struct list_head *mg_tasks_head;
910 void css_task_iter_start(struct cgroup_subsys_state *css,
911 struct css_task_iter *it);
912 struct task_struct *css_task_iter_next(struct css_task_iter *it);
913 void css_task_iter_end(struct css_task_iter *it);
915 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
916 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
918 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
919 struct cgroup_subsys *ss);
921 #else /* !CONFIG_CGROUPS */
923 static inline int cgroup_init_early(void) { return 0; }
924 static inline int cgroup_init(void) { return 0; }
925 static inline void cgroup_fork(struct task_struct *p) {}
926 static inline void cgroup_post_fork(struct task_struct *p) {}
927 static inline void cgroup_exit(struct task_struct *p) {}
929 static inline int cgroupstats_build(struct cgroupstats *stats,
930 struct dentry *dentry)
935 /* No cgroups - nothing to do */
936 static inline int cgroup_attach_task_all(struct task_struct *from,
937 struct task_struct *t)
942 #endif /* !CONFIG_CGROUPS */
944 #endif /* _LINUX_CGROUP_H */