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>
51 /* Per-subsystem/per-cgroup state maintained by the system. */
52 struct cgroup_subsys_state {
53 /* the cgroup that this css is attached to */
54 struct cgroup *cgroup;
56 /* the cgroup subsystem that this css is attached to */
57 struct cgroup_subsys *ss;
59 /* reference count - access via css_[try]get() and css_put() */
60 struct percpu_ref refcnt;
63 struct cgroup_subsys_state *parent;
66 * Subsys-unique ID. 0 is unused and root is always 1. The
67 * matching css can be looked up using css_from_id().
73 /* percpu_ref killing and RCU release */
74 struct rcu_head rcu_head;
75 struct work_struct destroy_work;
78 /* bits in struct cgroup_subsys_state flags field */
80 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
84 * css_get - obtain a reference on the specified css
87 * The caller must already have a reference.
89 static inline void css_get(struct cgroup_subsys_state *css)
91 percpu_ref_get(&css->refcnt);
95 * css_tryget_online - try to obtain a reference on the specified css if online
98 * Obtain a reference on @css if it's online. The caller naturally needs
99 * to ensure that @css is accessible but doesn't have to be holding a
100 * reference on it - IOW, RCU protected access is good enough for this
101 * function. Returns %true if a reference count was successfully obtained;
104 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
106 return percpu_ref_tryget_live(&css->refcnt);
110 * css_put - put a css reference
113 * Put a reference obtained via css_get() and css_tryget_online().
115 static inline void css_put(struct cgroup_subsys_state *css)
117 percpu_ref_put(&css->refcnt);
120 /* bits in struct cgroup flags field */
122 /* Control Group is dead */
125 * Control Group has previously had a child cgroup or a task,
126 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
129 /* Control Group requires release notifications to userspace */
130 CGRP_NOTIFY_ON_RELEASE,
132 * Clone the parent's configuration when creating a new child
133 * cpuset cgroup. For historical reasons, this option can be
134 * specified at mount time and thus is implemented here.
136 CGRP_CPUSET_CLONE_CHILDREN,
140 /* self css with NULL ->ss, points back to this cgroup */
141 struct cgroup_subsys_state self;
143 unsigned long flags; /* "unsigned long" so bitops work */
146 * idr allocated in-hierarchy ID.
148 * ID 0 is not used, the ID of the root cgroup is always 1, and a
149 * new cgroup will be assigned with a smallest available ID.
151 * Allocating/Removing ID must be protected by cgroup_mutex.
156 * If this cgroup contains any tasks, it contributes one to
157 * populated_cnt. All children with non-zero popuplated_cnt of
158 * their own contribute one. The count is zero iff there's no task
159 * in this cgroup or its subtree.
166 * We link our 'sibling' struct into our parent's 'children'.
167 * Our children link their 'sibling' into our 'children'.
169 struct list_head sibling; /* my parent's children */
170 struct list_head children; /* my children */
172 struct cgroup *parent; /* my parent */
173 struct kernfs_node *kn; /* cgroup kernfs entry */
174 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
177 * Monotonically increasing unique serial number which defines a
178 * uniform order among all cgroups. It's guaranteed that all
179 * ->children lists are in the ascending order of ->serial_nr.
180 * It's used to allow interrupting and resuming iterations.
184 /* the bitmask of subsystems enabled on the child cgroups */
185 unsigned int child_subsys_mask;
187 /* Private pointers for each registered subsystem */
188 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
190 struct cgroup_root *root;
193 * List of cgrp_cset_links pointing at css_sets with tasks in this
194 * cgroup. Protected by css_set_lock.
196 struct list_head cset_links;
199 * On the default hierarchy, a css_set for a cgroup with some
200 * susbsys disabled will point to css's which are associated with
201 * the closest ancestor which has the subsys enabled. The
202 * following lists all css_sets which point to this cgroup's css
203 * for the given subsystem.
205 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
208 * Linked list running through all cgroups that can
209 * potentially be reaped by the release agent. Protected by
212 struct list_head release_list;
215 * list of pidlists, up to two for each namespace (one for procs, one
216 * for tasks); created on demand.
218 struct list_head pidlists;
219 struct mutex pidlist_mutex;
221 /* For css percpu_ref killing and RCU-protected deletion */
222 struct rcu_head rcu_head;
223 struct work_struct destroy_work;
225 /* used to wait for offlining of csses */
226 wait_queue_head_t offline_waitq;
229 #define MAX_CGROUP_ROOT_NAMELEN 64
231 /* cgroup_root->flags */
234 * Unfortunately, cgroup core and various controllers are riddled
235 * with idiosyncrasies and pointless options. The following flag,
236 * when set, will force sane behavior - some options are forced on,
237 * others are disallowed, and some controllers will change their
238 * hierarchical or other behaviors.
240 * The set of behaviors affected by this flag are still being
241 * determined and developed and the mount option for this flag is
242 * prefixed with __DEVEL__. The prefix will be dropped once we
243 * reach the point where all behaviors are compatible with the
244 * planned unified hierarchy, which will automatically turn on this
247 * The followings are the behaviors currently affected this flag.
249 * - Mount options "noprefix", "xattr", "clone_children",
250 * "release_agent" and "name" are disallowed.
252 * - When mounting an existing superblock, mount options should
255 * - Remount is disallowed.
257 * - rename(2) is disallowed.
259 * - "tasks" is removed. Everything should be at process
260 * granularity. Use "cgroup.procs" instead.
262 * - "cgroup.procs" is not sorted. pids will be unique unless they
263 * got recycled inbetween reads.
265 * - "release_agent" and "notify_on_release" are removed.
266 * Replacement notification mechanism will be implemented.
268 * - "cgroup.clone_children" is removed.
270 * - "cgroup.subtree_populated" is available. Its value is 0 if
271 * the cgroup and its descendants contain no task; otherwise, 1.
272 * The file also generates kernfs notification which can be
273 * monitored through poll and [di]notify when the value of the
276 * - If mount is requested with sane_behavior but without any
277 * subsystem, the default unified hierarchy is mounted.
279 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
280 * and take masks of ancestors with non-empty cpus/mems, instead of
281 * being moved to an ancestor.
283 * - cpuset: a task can be moved into an empty cpuset, and again it
284 * takes masks of ancestors.
286 * - memcg: use_hierarchy is on by default and the cgroup file for
287 * the flag is not created.
289 * - blkcg: blk-throttle becomes properly hierarchical.
291 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
293 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
294 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
296 /* mount options live below bit 16 */
297 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
301 * A cgroup_root represents the root of a cgroup hierarchy, and may be
302 * associated with a kernfs_root to form an active hierarchy. This is
303 * internal to cgroup core. Don't access directly from controllers.
306 struct kernfs_root *kf_root;
308 /* The bitmask of subsystems attached to this hierarchy */
309 unsigned int subsys_mask;
311 /* Unique id for this hierarchy. */
314 /* The root cgroup. Root is destroyed on its release. */
317 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
320 /* A list running through the active hierarchies */
321 struct list_head root_list;
323 /* Hierarchy-specific flags */
326 /* IDs for cgroups in this hierarchy */
327 struct idr cgroup_idr;
329 /* The path to use for release notifications. */
330 char release_agent_path[PATH_MAX];
332 /* The name for this hierarchy - may be empty */
333 char name[MAX_CGROUP_ROOT_NAMELEN];
337 * A css_set is a structure holding pointers to a set of
338 * cgroup_subsys_state objects. This saves space in the task struct
339 * object and speeds up fork()/exit(), since a single inc/dec and a
340 * list_add()/del() can bump the reference count on the entire cgroup
346 /* Reference count */
350 * List running through all cgroup groups in the same hash
351 * slot. Protected by css_set_lock
353 struct hlist_node hlist;
356 * Lists running through all tasks using this cgroup group.
357 * mg_tasks lists tasks which belong to this cset but are in the
358 * process of being migrated out or in. Protected by
359 * css_set_rwsem, but, during migration, once tasks are moved to
360 * mg_tasks, it can be read safely while holding cgroup_mutex.
362 struct list_head tasks;
363 struct list_head mg_tasks;
366 * List of cgrp_cset_links pointing at cgroups referenced from this
367 * css_set. Protected by css_set_lock.
369 struct list_head cgrp_links;
371 /* the default cgroup associated with this css_set */
372 struct cgroup *dfl_cgrp;
375 * Set of subsystem states, one for each subsystem. This array is
376 * immutable after creation apart from the init_css_set during
377 * subsystem registration (at boot time).
379 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
382 * List of csets participating in the on-going migration either as
383 * source or destination. Protected by cgroup_mutex.
385 struct list_head mg_preload_node;
386 struct list_head mg_node;
389 * If this cset is acting as the source of migration the following
390 * two fields are set. mg_src_cgrp is the source cgroup of the
391 * on-going migration and mg_dst_cset is the destination cset the
392 * target tasks on this cset should be migrated to. Protected by
395 struct cgroup *mg_src_cgrp;
396 struct css_set *mg_dst_cset;
399 * On the default hierarhcy, ->subsys[ssid] may point to a css
400 * attached to an ancestor instead of the cgroup this css_set is
401 * associated with. The following node is anchored at
402 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
403 * iterate through all css's attached to a given cgroup.
405 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
407 /* For RCU-protected deletion */
408 struct rcu_head rcu_head;
412 * struct cftype: handler definitions for cgroup control files
414 * When reading/writing to a file:
415 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
416 * - the 'cftype' of the file is file->f_dentry->d_fsdata
421 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
422 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
423 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
424 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
425 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
428 #define MAX_CFTYPE_NAME 64
432 * By convention, the name should begin with the name of the
433 * subsystem, followed by a period. Zero length string indicates
434 * end of cftype array.
436 char name[MAX_CFTYPE_NAME];
439 * If not 0, file mode is set to this value, otherwise it will
440 * be figured out automatically
445 * The maximum length of string, excluding trailing nul, that can
446 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
448 size_t max_write_len;
454 * Fields used for internal bookkeeping. Initialized automatically
455 * during registration.
457 struct cgroup_subsys *ss; /* NULL for cgroup core files */
458 struct list_head node; /* anchored at ss->cfts */
459 struct kernfs_ops *kf_ops;
462 * read_u64() is a shortcut for the common case of returning a
463 * single integer. Use it in place of read()
465 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
467 * read_s64() is a signed version of read_u64()
469 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
471 /* generic seq_file read interface */
472 int (*seq_show)(struct seq_file *sf, void *v);
474 /* optional ops, implement all or none */
475 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
476 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
477 void (*seq_stop)(struct seq_file *sf, void *v);
480 * write_u64() is a shortcut for the common case of accepting
481 * a single integer (as parsed by simple_strtoull) from
482 * userspace. Use in place of write(); return 0 or error.
484 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
487 * write_s64() is a signed version of write_u64()
489 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
493 * write() is the generic write callback which maps directly to
494 * kernfs write operation and overrides all other operations.
495 * Maximum write size is determined by ->max_write_len. Use
496 * of_css/cft() to access the associated css and cft.
498 ssize_t (*write)(struct kernfs_open_file *of,
499 char *buf, size_t nbytes, loff_t off);
501 #ifdef CONFIG_DEBUG_LOCK_ALLOC
502 struct lock_class_key lockdep_key;
506 extern struct cgroup_root cgrp_dfl_root;
507 extern struct css_set init_css_set;
509 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
511 return cgrp->root == &cgrp_dfl_root;
515 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
516 * function can be called as long as @cgrp is accessible.
518 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
520 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
523 /* no synchronization, the result can only be used as a hint */
524 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
526 return !list_empty(&cgrp->cset_links);
529 /* returns ino associated with a cgroup, 0 indicates unmounted root */
530 static inline ino_t cgroup_ino(struct cgroup *cgrp)
533 return cgrp->kn->ino;
538 /* cft/css accessors for cftype->write() operation */
539 static inline struct cftype *of_cft(struct kernfs_open_file *of)
544 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
546 /* cft/css accessors for cftype->seq_*() operations */
547 static inline struct cftype *seq_cft(struct seq_file *seq)
549 return of_cft(seq->private);
552 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
554 return of_css(seq->private);
558 * Name / path handling functions. All are thin wrappers around the kernfs
559 * counterparts and can be called under any context.
562 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
564 return kernfs_name(cgrp->kn, buf, buflen);
567 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
570 return kernfs_path(cgrp->kn, buf, buflen);
573 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
575 pr_cont_kernfs_name(cgrp->kn);
578 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
580 pr_cont_kernfs_path(cgrp->kn);
583 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
585 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
586 int cgroup_rm_cftypes(struct cftype *cfts);
588 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
591 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
594 struct cgroup_taskset;
595 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
596 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
599 * cgroup_taskset_for_each - iterate cgroup_taskset
600 * @task: the loop cursor
601 * @tset: taskset to iterate
603 #define cgroup_taskset_for_each(task, tset) \
604 for ((task) = cgroup_taskset_first((tset)); (task); \
605 (task) = cgroup_taskset_next((tset)))
608 * Control Group subsystem type.
609 * See Documentation/cgroups/cgroups.txt for details
612 struct cgroup_subsys {
613 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
614 int (*css_online)(struct cgroup_subsys_state *css);
615 void (*css_offline)(struct cgroup_subsys_state *css);
616 void (*css_free)(struct cgroup_subsys_state *css);
618 int (*can_attach)(struct cgroup_subsys_state *css,
619 struct cgroup_taskset *tset);
620 void (*cancel_attach)(struct cgroup_subsys_state *css,
621 struct cgroup_taskset *tset);
622 void (*attach)(struct cgroup_subsys_state *css,
623 struct cgroup_taskset *tset);
624 void (*fork)(struct task_struct *task);
625 void (*exit)(struct cgroup_subsys_state *css,
626 struct cgroup_subsys_state *old_css,
627 struct task_struct *task);
628 void (*bind)(struct cgroup_subsys_state *root_css);
634 * If %false, this subsystem is properly hierarchical -
635 * configuration, resource accounting and restriction on a parent
636 * cgroup cover those of its children. If %true, hierarchy support
637 * is broken in some ways - some subsystems ignore hierarchy
638 * completely while others are only implemented half-way.
640 * It's now disallowed to create nested cgroups if the subsystem is
641 * broken and cgroup core will emit a warning message on such
642 * cases. Eventually, all subsystems will be made properly
643 * hierarchical and this will go away.
645 bool broken_hierarchy;
646 bool warned_broken_hierarchy;
648 /* the following two fields are initialized automtically during boot */
650 #define MAX_CGROUP_TYPE_NAMELEN 32
653 /* link to parent, protected by cgroup_lock() */
654 struct cgroup_root *root;
656 /* idr for css->id */
660 * List of cftypes. Each entry is the first entry of an array
661 * terminated by zero length name.
663 struct list_head cfts;
665 /* base cftypes, automatically registered with subsys itself */
666 struct cftype *base_cftypes;
669 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
670 #include <linux/cgroup_subsys.h>
674 * css_parent - find the parent css
675 * @css: the target cgroup_subsys_state
677 * Return the parent css of @css. This function is guaranteed to return
678 * non-NULL parent as long as @css isn't the root.
681 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
687 * task_css_set_check - obtain a task's css_set with extra access conditions
688 * @task: the task to obtain css_set for
689 * @__c: extra condition expression to be passed to rcu_dereference_check()
691 * A task's css_set is RCU protected, initialized and exited while holding
692 * task_lock(), and can only be modified while holding both cgroup_mutex
693 * and task_lock() while the task is alive. This macro verifies that the
694 * caller is inside proper critical section and returns @task's css_set.
696 * The caller can also specify additional allowed conditions via @__c, such
697 * as locks used during the cgroup_subsys::attach() methods.
699 #ifdef CONFIG_PROVE_RCU
700 extern struct mutex cgroup_mutex;
701 extern struct rw_semaphore css_set_rwsem;
702 #define task_css_set_check(task, __c) \
703 rcu_dereference_check((task)->cgroups, \
704 lockdep_is_held(&cgroup_mutex) || \
705 lockdep_is_held(&css_set_rwsem) || \
706 ((task)->flags & PF_EXITING) || (__c))
708 #define task_css_set_check(task, __c) \
709 rcu_dereference((task)->cgroups)
713 * task_css_check - obtain css for (task, subsys) w/ extra access conds
714 * @task: the target task
715 * @subsys_id: the target subsystem ID
716 * @__c: extra condition expression to be passed to rcu_dereference_check()
718 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
719 * synchronization rules are the same as task_css_set_check().
721 #define task_css_check(task, subsys_id, __c) \
722 task_css_set_check((task), (__c))->subsys[(subsys_id)]
725 * task_css_set - obtain a task's css_set
726 * @task: the task to obtain css_set for
728 * See task_css_set_check().
730 static inline struct css_set *task_css_set(struct task_struct *task)
732 return task_css_set_check(task, false);
736 * task_css - obtain css for (task, subsys)
737 * @task: the target task
738 * @subsys_id: the target subsystem ID
740 * See task_css_check().
742 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
745 return task_css_check(task, subsys_id, false);
749 * task_css_is_root - test whether a task belongs to the root css
750 * @task: the target task
751 * @subsys_id: the target subsystem ID
753 * Test whether @task belongs to the root css on the specified subsystem.
754 * May be invoked in any context.
756 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
758 return task_css_check(task, subsys_id, true) ==
759 init_css_set.subsys[subsys_id];
762 static inline struct cgroup *task_cgroup(struct task_struct *task,
765 return task_css(task, subsys_id)->cgroup;
768 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
769 struct cgroup_subsys_state *parent);
771 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
774 * css_for_each_child - iterate through children of a css
775 * @pos: the css * to use as the loop cursor
776 * @parent: css whose children to walk
778 * Walk @parent's children. Must be called under rcu_read_lock(). A child
779 * css which hasn't finished ->css_online() or already has finished
780 * ->css_offline() may show up during traversal and it's each subsystem's
781 * responsibility to verify that each @pos is alive.
783 * If a subsystem synchronizes against the parent in its ->css_online() and
784 * before starting iterating, a css which finished ->css_online() is
785 * guaranteed to be visible in the future iterations.
787 * It is allowed to temporarily drop RCU read lock during iteration. The
788 * caller is responsible for ensuring that @pos remains accessible until
789 * the start of the next iteration by, for example, bumping the css refcnt.
791 #define css_for_each_child(pos, parent) \
792 for ((pos) = css_next_child(NULL, (parent)); (pos); \
793 (pos) = css_next_child((pos), (parent)))
795 struct cgroup_subsys_state *
796 css_next_descendant_pre(struct cgroup_subsys_state *pos,
797 struct cgroup_subsys_state *css);
799 struct cgroup_subsys_state *
800 css_rightmost_descendant(struct cgroup_subsys_state *pos);
803 * css_for_each_descendant_pre - pre-order walk of a css's descendants
804 * @pos: the css * to use as the loop cursor
805 * @root: css whose descendants to walk
807 * Walk @root's descendants. @root is included in the iteration and the
808 * first node to be visited. Must be called under rcu_read_lock(). A
809 * descendant css which hasn't finished ->css_online() or already has
810 * finished ->css_offline() may show up during traversal and it's each
811 * subsystem's responsibility to verify that each @pos is alive.
813 * If a subsystem synchronizes against the parent in its ->css_online() and
814 * before starting iterating, and synchronizes against @pos on each
815 * iteration, any descendant css which finished ->css_online() is
816 * guaranteed to be visible in the future iterations.
818 * In other words, the following guarantees that a descendant can't escape
819 * state updates of its ancestors.
823 * Lock @css's parent and @css;
824 * Inherit state from the parent;
828 * my_update_state(@css)
830 * css_for_each_descendant_pre(@pos, @css) {
833 * Update @css's state;
835 * Verify @pos is alive and inherit state from its parent;
840 * As long as the inheriting step, including checking the parent state, is
841 * enclosed inside @pos locking, double-locking the parent isn't necessary
842 * while inheriting. The state update to the parent is guaranteed to be
843 * visible by walking order and, as long as inheriting operations to the
844 * same @pos are atomic to each other, multiple updates racing each other
845 * still result in the correct state. It's guaranateed that at least one
846 * inheritance happens for any css after the latest update to its parent.
848 * If checking parent's state requires locking the parent, each inheriting
849 * iteration should lock and unlock both @pos->parent and @pos.
851 * Alternatively, a subsystem may choose to use a single global lock to
852 * synchronize ->css_online() and ->css_offline() against tree-walking
855 * It is allowed to temporarily drop RCU read lock during iteration. The
856 * caller is responsible for ensuring that @pos remains accessible until
857 * the start of the next iteration by, for example, bumping the css refcnt.
859 #define css_for_each_descendant_pre(pos, css) \
860 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
861 (pos) = css_next_descendant_pre((pos), (css)))
863 struct cgroup_subsys_state *
864 css_next_descendant_post(struct cgroup_subsys_state *pos,
865 struct cgroup_subsys_state *css);
868 * css_for_each_descendant_post - post-order walk of a css's descendants
869 * @pos: the css * to use as the loop cursor
870 * @css: css whose descendants to walk
872 * Similar to css_for_each_descendant_pre() but performs post-order
873 * traversal instead. @root is included in the iteration and the last
874 * node to be visited. Note that the walk visibility guarantee described
875 * in pre-order walk doesn't apply the same to post-order walks.
877 #define css_for_each_descendant_post(pos, css) \
878 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
879 (pos) = css_next_descendant_post((pos), (css)))
881 /* A css_task_iter should be treated as an opaque object */
882 struct css_task_iter {
883 struct cgroup_subsys *ss;
885 struct list_head *cset_pos;
886 struct list_head *cset_head;
888 struct list_head *task_pos;
889 struct list_head *tasks_head;
890 struct list_head *mg_tasks_head;
893 void css_task_iter_start(struct cgroup_subsys_state *css,
894 struct css_task_iter *it);
895 struct task_struct *css_task_iter_next(struct css_task_iter *it);
896 void css_task_iter_end(struct css_task_iter *it);
898 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
899 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
901 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
902 struct cgroup_subsys *ss);
904 #else /* !CONFIG_CGROUPS */
906 static inline int cgroup_init_early(void) { return 0; }
907 static inline int cgroup_init(void) { return 0; }
908 static inline void cgroup_fork(struct task_struct *p) {}
909 static inline void cgroup_post_fork(struct task_struct *p) {}
910 static inline void cgroup_exit(struct task_struct *p) {}
912 static inline int cgroupstats_build(struct cgroupstats *stats,
913 struct dentry *dentry)
918 /* No cgroups - nothing to do */
919 static inline int cgroup_attach_task_all(struct task_struct *from,
920 struct task_struct *t)
925 #endif /* !CONFIG_CGROUPS */
927 #endif /* _LINUX_CGROUP_H */