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_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
81 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
85 * css_get - obtain a reference on the specified css
88 * The caller must already have a reference.
90 static inline void css_get(struct cgroup_subsys_state *css)
92 /* We don't need to reference count the root state */
93 if (!(css->flags & CSS_ROOT))
94 percpu_ref_get(&css->refcnt);
98 * css_tryget_online - try to obtain a reference on the specified css if online
101 * Obtain a reference on @css if it's online. The caller naturally needs
102 * to ensure that @css is accessible but doesn't have to be holding a
103 * reference on it - IOW, RCU protected access is good enough for this
104 * function. Returns %true if a reference count was successfully obtained;
107 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
109 if (css->flags & CSS_ROOT)
111 return percpu_ref_tryget_live(&css->refcnt);
115 * css_put - put a css reference
118 * Put a reference obtained via css_get() and css_tryget_online().
120 static inline void css_put(struct cgroup_subsys_state *css)
122 if (!(css->flags & CSS_ROOT))
123 percpu_ref_put(&css->refcnt);
126 /* bits in struct cgroup flags field */
128 /* Control Group is dead */
131 * Control Group has previously had a child cgroup or a task,
132 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
135 /* Control Group requires release notifications to userspace */
136 CGRP_NOTIFY_ON_RELEASE,
138 * Clone the parent's configuration when creating a new child
139 * cpuset cgroup. For historical reasons, this option can be
140 * specified at mount time and thus is implemented here.
142 CGRP_CPUSET_CLONE_CHILDREN,
146 unsigned long flags; /* "unsigned long" so bitops work */
149 * idr allocated in-hierarchy ID.
151 * ID 0 is not used, the ID of the root cgroup is always 1, and a
152 * new cgroup will be assigned with a smallest available ID.
154 * Allocating/Removing ID must be protected by cgroup_mutex.
158 /* the number of attached css's */
162 * If this cgroup contains any tasks, it contributes one to
163 * populated_cnt. All children with non-zero popuplated_cnt of
164 * their own contribute one. The count is zero iff there's no task
165 * in this cgroup or its subtree.
172 * We link our 'sibling' struct into our parent's 'children'.
173 * Our children link their 'sibling' into our 'children'.
175 struct list_head sibling; /* my parent's children */
176 struct list_head children; /* my children */
178 struct cgroup *parent; /* my parent */
179 struct kernfs_node *kn; /* cgroup kernfs entry */
180 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
183 * Monotonically increasing unique serial number which defines a
184 * uniform order among all cgroups. It's guaranteed that all
185 * ->children lists are in the ascending order of ->serial_nr.
186 * It's used to allow interrupting and resuming iterations.
190 /* the bitmask of subsystems enabled on the child cgroups */
191 unsigned int child_subsys_mask;
193 /* Private pointers for each registered subsystem */
194 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
196 struct cgroup_root *root;
199 * List of cgrp_cset_links pointing at css_sets with tasks in this
200 * cgroup. Protected by css_set_lock.
202 struct list_head cset_links;
205 * On the default hierarchy, a css_set for a cgroup with some
206 * susbsys disabled will point to css's which are associated with
207 * the closest ancestor which has the subsys enabled. The
208 * following lists all css_sets which point to this cgroup's css
209 * for the given subsystem.
211 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
214 * Linked list running through all cgroups that can
215 * potentially be reaped by the release agent. Protected by
218 struct list_head release_list;
221 * list of pidlists, up to two for each namespace (one for procs, one
222 * for tasks); created on demand.
224 struct list_head pidlists;
225 struct mutex pidlist_mutex;
227 /* dummy css with NULL ->ss, points back to this cgroup */
228 struct cgroup_subsys_state dummy_css;
230 /* For css percpu_ref killing and RCU-protected deletion */
231 struct rcu_head rcu_head;
232 struct work_struct destroy_work;
234 /* used to wait for offlining of csses */
235 wait_queue_head_t offline_waitq;
238 #define MAX_CGROUP_ROOT_NAMELEN 64
240 /* cgroup_root->flags */
243 * Unfortunately, cgroup core and various controllers are riddled
244 * with idiosyncrasies and pointless options. The following flag,
245 * when set, will force sane behavior - some options are forced on,
246 * others are disallowed, and some controllers will change their
247 * hierarchical or other behaviors.
249 * The set of behaviors affected by this flag are still being
250 * determined and developed and the mount option for this flag is
251 * prefixed with __DEVEL__. The prefix will be dropped once we
252 * reach the point where all behaviors are compatible with the
253 * planned unified hierarchy, which will automatically turn on this
256 * The followings are the behaviors currently affected this flag.
258 * - Mount options "noprefix", "xattr", "clone_children",
259 * "release_agent" and "name" are disallowed.
261 * - When mounting an existing superblock, mount options should
264 * - Remount is disallowed.
266 * - rename(2) is disallowed.
268 * - "tasks" is removed. Everything should be at process
269 * granularity. Use "cgroup.procs" instead.
271 * - "cgroup.procs" is not sorted. pids will be unique unless they
272 * got recycled inbetween reads.
274 * - "release_agent" and "notify_on_release" are removed.
275 * Replacement notification mechanism will be implemented.
277 * - "cgroup.clone_children" is removed.
279 * - "cgroup.subtree_populated" is available. Its value is 0 if
280 * the cgroup and its descendants contain no task; otherwise, 1.
281 * The file also generates kernfs notification which can be
282 * monitored through poll and [di]notify when the value of the
285 * - If mount is requested with sane_behavior but without any
286 * subsystem, the default unified hierarchy is mounted.
288 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
289 * and take masks of ancestors with non-empty cpus/mems, instead of
290 * being moved to an ancestor.
292 * - cpuset: a task can be moved into an empty cpuset, and again it
293 * takes masks of ancestors.
295 * - memcg: use_hierarchy is on by default and the cgroup file for
296 * the flag is not created.
298 * - blkcg: blk-throttle becomes properly hierarchical.
300 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
302 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
303 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
305 /* mount options live below bit 16 */
306 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
310 * A cgroup_root represents the root of a cgroup hierarchy, and may be
311 * associated with a kernfs_root to form an active hierarchy. This is
312 * internal to cgroup core. Don't access directly from controllers.
315 struct kernfs_root *kf_root;
317 /* The bitmask of subsystems attached to this hierarchy */
318 unsigned int subsys_mask;
320 /* Unique id for this hierarchy. */
323 /* The root cgroup. Root is destroyed on its release. */
326 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
329 /* A list running through the active hierarchies */
330 struct list_head root_list;
332 /* Hierarchy-specific flags */
335 /* IDs for cgroups in this hierarchy */
336 struct idr cgroup_idr;
338 /* The path to use for release notifications. */
339 char release_agent_path[PATH_MAX];
341 /* The name for this hierarchy - may be empty */
342 char name[MAX_CGROUP_ROOT_NAMELEN];
346 * A css_set is a structure holding pointers to a set of
347 * cgroup_subsys_state objects. This saves space in the task struct
348 * object and speeds up fork()/exit(), since a single inc/dec and a
349 * list_add()/del() can bump the reference count on the entire cgroup
355 /* Reference count */
359 * List running through all cgroup groups in the same hash
360 * slot. Protected by css_set_lock
362 struct hlist_node hlist;
365 * Lists running through all tasks using this cgroup group.
366 * mg_tasks lists tasks which belong to this cset but are in the
367 * process of being migrated out or in. Protected by
368 * css_set_rwsem, but, during migration, once tasks are moved to
369 * mg_tasks, it can be read safely while holding cgroup_mutex.
371 struct list_head tasks;
372 struct list_head mg_tasks;
375 * List of cgrp_cset_links pointing at cgroups referenced from this
376 * css_set. Protected by css_set_lock.
378 struct list_head cgrp_links;
380 /* the default cgroup associated with this css_set */
381 struct cgroup *dfl_cgrp;
384 * Set of subsystem states, one for each subsystem. This array is
385 * immutable after creation apart from the init_css_set during
386 * subsystem registration (at boot time).
388 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
391 * List of csets participating in the on-going migration either as
392 * source or destination. Protected by cgroup_mutex.
394 struct list_head mg_preload_node;
395 struct list_head mg_node;
398 * If this cset is acting as the source of migration the following
399 * two fields are set. mg_src_cgrp is the source cgroup of the
400 * on-going migration and mg_dst_cset is the destination cset the
401 * target tasks on this cset should be migrated to. Protected by
404 struct cgroup *mg_src_cgrp;
405 struct css_set *mg_dst_cset;
408 * On the default hierarhcy, ->subsys[ssid] may point to a css
409 * attached to an ancestor instead of the cgroup this css_set is
410 * associated with. The following node is anchored at
411 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
412 * iterate through all css's attached to a given cgroup.
414 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
416 /* For RCU-protected deletion */
417 struct rcu_head rcu_head;
421 * struct cftype: handler definitions for cgroup control files
423 * When reading/writing to a file:
424 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
425 * - the 'cftype' of the file is file->f_dentry->d_fsdata
430 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
431 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
432 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
433 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
434 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
437 #define MAX_CFTYPE_NAME 64
441 * By convention, the name should begin with the name of the
442 * subsystem, followed by a period. Zero length string indicates
443 * end of cftype array.
445 char name[MAX_CFTYPE_NAME];
448 * If not 0, file mode is set to this value, otherwise it will
449 * be figured out automatically
454 * The maximum length of string, excluding trailing nul, that can
455 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
457 size_t max_write_len;
463 * Fields used for internal bookkeeping. Initialized automatically
464 * during registration.
466 struct cgroup_subsys *ss; /* NULL for cgroup core files */
467 struct list_head node; /* anchored at ss->cfts */
468 struct kernfs_ops *kf_ops;
471 * read_u64() is a shortcut for the common case of returning a
472 * single integer. Use it in place of read()
474 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
476 * read_s64() is a signed version of read_u64()
478 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
480 /* generic seq_file read interface */
481 int (*seq_show)(struct seq_file *sf, void *v);
483 /* optional ops, implement all or none */
484 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
485 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
486 void (*seq_stop)(struct seq_file *sf, void *v);
489 * write_u64() is a shortcut for the common case of accepting
490 * a single integer (as parsed by simple_strtoull) from
491 * userspace. Use in place of write(); return 0 or error.
493 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
496 * write_s64() is a signed version of write_u64()
498 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
502 * write() is the generic write callback which maps directly to
503 * kernfs write operation and overrides all other operations.
504 * Maximum write size is determined by ->max_write_len. Use
505 * of_css/cft() to access the associated css and cft.
507 ssize_t (*write)(struct kernfs_open_file *of,
508 char *buf, size_t nbytes, loff_t off);
510 #ifdef CONFIG_DEBUG_LOCK_ALLOC
511 struct lock_class_key lockdep_key;
515 extern struct cgroup_root cgrp_dfl_root;
516 extern struct css_set init_css_set;
518 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
520 return cgrp->root == &cgrp_dfl_root;
524 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
525 * function can be called as long as @cgrp is accessible.
527 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
529 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
532 /* no synchronization, the result can only be used as a hint */
533 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
535 return !list_empty(&cgrp->cset_links);
538 /* returns ino associated with a cgroup, 0 indicates unmounted root */
539 static inline ino_t cgroup_ino(struct cgroup *cgrp)
542 return cgrp->kn->ino;
547 /* cft/css accessors for cftype->write() operation */
548 static inline struct cftype *of_cft(struct kernfs_open_file *of)
553 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
555 /* cft/css accessors for cftype->seq_*() operations */
556 static inline struct cftype *seq_cft(struct seq_file *seq)
558 return of_cft(seq->private);
561 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
563 return of_css(seq->private);
567 * Name / path handling functions. All are thin wrappers around the kernfs
568 * counterparts and can be called under any context.
571 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
573 return kernfs_name(cgrp->kn, buf, buflen);
576 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
579 return kernfs_path(cgrp->kn, buf, buflen);
582 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
584 pr_cont_kernfs_name(cgrp->kn);
587 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
589 pr_cont_kernfs_path(cgrp->kn);
592 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
594 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
595 int cgroup_rm_cftypes(struct cftype *cfts);
597 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
600 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
603 struct cgroup_taskset;
604 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
605 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
608 * cgroup_taskset_for_each - iterate cgroup_taskset
609 * @task: the loop cursor
610 * @tset: taskset to iterate
612 #define cgroup_taskset_for_each(task, tset) \
613 for ((task) = cgroup_taskset_first((tset)); (task); \
614 (task) = cgroup_taskset_next((tset)))
617 * Control Group subsystem type.
618 * See Documentation/cgroups/cgroups.txt for details
621 struct cgroup_subsys {
622 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
623 int (*css_online)(struct cgroup_subsys_state *css);
624 void (*css_offline)(struct cgroup_subsys_state *css);
625 void (*css_free)(struct cgroup_subsys_state *css);
627 int (*can_attach)(struct cgroup_subsys_state *css,
628 struct cgroup_taskset *tset);
629 void (*cancel_attach)(struct cgroup_subsys_state *css,
630 struct cgroup_taskset *tset);
631 void (*attach)(struct cgroup_subsys_state *css,
632 struct cgroup_taskset *tset);
633 void (*fork)(struct task_struct *task);
634 void (*exit)(struct cgroup_subsys_state *css,
635 struct cgroup_subsys_state *old_css,
636 struct task_struct *task);
637 void (*bind)(struct cgroup_subsys_state *root_css);
643 * If %false, this subsystem is properly hierarchical -
644 * configuration, resource accounting and restriction on a parent
645 * cgroup cover those of its children. If %true, hierarchy support
646 * is broken in some ways - some subsystems ignore hierarchy
647 * completely while others are only implemented half-way.
649 * It's now disallowed to create nested cgroups if the subsystem is
650 * broken and cgroup core will emit a warning message on such
651 * cases. Eventually, all subsystems will be made properly
652 * hierarchical and this will go away.
654 bool broken_hierarchy;
655 bool warned_broken_hierarchy;
657 /* the following two fields are initialized automtically during boot */
659 #define MAX_CGROUP_TYPE_NAMELEN 32
662 /* link to parent, protected by cgroup_lock() */
663 struct cgroup_root *root;
665 /* idr for css->id */
669 * List of cftypes. Each entry is the first entry of an array
670 * terminated by zero length name.
672 struct list_head cfts;
674 /* base cftypes, automatically registered with subsys itself */
675 struct cftype *base_cftypes;
678 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
679 #include <linux/cgroup_subsys.h>
683 * css_parent - find the parent css
684 * @css: the target cgroup_subsys_state
686 * Return the parent css of @css. This function is guaranteed to return
687 * non-NULL parent as long as @css isn't the root.
690 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
696 * task_css_set_check - obtain a task's css_set with extra access conditions
697 * @task: the task to obtain css_set for
698 * @__c: extra condition expression to be passed to rcu_dereference_check()
700 * A task's css_set is RCU protected, initialized and exited while holding
701 * task_lock(), and can only be modified while holding both cgroup_mutex
702 * and task_lock() while the task is alive. This macro verifies that the
703 * caller is inside proper critical section and returns @task's css_set.
705 * The caller can also specify additional allowed conditions via @__c, such
706 * as locks used during the cgroup_subsys::attach() methods.
708 #ifdef CONFIG_PROVE_RCU
709 extern struct mutex cgroup_mutex;
710 extern struct rw_semaphore css_set_rwsem;
711 #define task_css_set_check(task, __c) \
712 rcu_dereference_check((task)->cgroups, \
713 lockdep_is_held(&cgroup_mutex) || \
714 lockdep_is_held(&css_set_rwsem) || \
715 ((task)->flags & PF_EXITING) || (__c))
717 #define task_css_set_check(task, __c) \
718 rcu_dereference((task)->cgroups)
722 * task_css_check - obtain css for (task, subsys) w/ extra access conds
723 * @task: the target task
724 * @subsys_id: the target subsystem ID
725 * @__c: extra condition expression to be passed to rcu_dereference_check()
727 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
728 * synchronization rules are the same as task_css_set_check().
730 #define task_css_check(task, subsys_id, __c) \
731 task_css_set_check((task), (__c))->subsys[(subsys_id)]
734 * task_css_set - obtain a task's css_set
735 * @task: the task to obtain css_set for
737 * See task_css_set_check().
739 static inline struct css_set *task_css_set(struct task_struct *task)
741 return task_css_set_check(task, false);
745 * task_css - obtain css for (task, subsys)
746 * @task: the target task
747 * @subsys_id: the target subsystem ID
749 * See task_css_check().
751 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
754 return task_css_check(task, subsys_id, false);
758 * task_css_is_root - test whether a task belongs to the root css
759 * @task: the target task
760 * @subsys_id: the target subsystem ID
762 * Test whether @task belongs to the root css on the specified subsystem.
763 * May be invoked in any context.
765 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
767 return task_css_check(task, subsys_id, true) ==
768 init_css_set.subsys[subsys_id];
771 static inline struct cgroup *task_cgroup(struct task_struct *task,
774 return task_css(task, subsys_id)->cgroup;
777 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
778 struct cgroup_subsys_state *parent);
780 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
783 * css_for_each_child - iterate through children of a css
784 * @pos: the css * to use as the loop cursor
785 * @parent: css whose children to walk
787 * Walk @parent's children. Must be called under rcu_read_lock(). A child
788 * css which hasn't finished ->css_online() or already has finished
789 * ->css_offline() may show up during traversal and it's each subsystem's
790 * responsibility to verify that each @pos is alive.
792 * If a subsystem synchronizes against the parent in its ->css_online() and
793 * before starting iterating, a css which finished ->css_online() is
794 * guaranteed to be visible in the future iterations.
796 * It is allowed to temporarily drop RCU read lock during iteration. The
797 * caller is responsible for ensuring that @pos remains accessible until
798 * the start of the next iteration by, for example, bumping the css refcnt.
800 #define css_for_each_child(pos, parent) \
801 for ((pos) = css_next_child(NULL, (parent)); (pos); \
802 (pos) = css_next_child((pos), (parent)))
804 struct cgroup_subsys_state *
805 css_next_descendant_pre(struct cgroup_subsys_state *pos,
806 struct cgroup_subsys_state *css);
808 struct cgroup_subsys_state *
809 css_rightmost_descendant(struct cgroup_subsys_state *pos);
812 * css_for_each_descendant_pre - pre-order walk of a css's descendants
813 * @pos: the css * to use as the loop cursor
814 * @root: css whose descendants to walk
816 * Walk @root's descendants. @root is included in the iteration and the
817 * first node to be visited. Must be called under rcu_read_lock(). A
818 * descendant css which hasn't finished ->css_online() or already has
819 * finished ->css_offline() may show up during traversal and it's each
820 * subsystem's responsibility to verify that each @pos is alive.
822 * If a subsystem synchronizes against the parent in its ->css_online() and
823 * before starting iterating, and synchronizes against @pos on each
824 * iteration, any descendant css which finished ->css_online() is
825 * guaranteed to be visible in the future iterations.
827 * In other words, the following guarantees that a descendant can't escape
828 * state updates of its ancestors.
832 * Lock @css's parent and @css;
833 * Inherit state from the parent;
837 * my_update_state(@css)
839 * css_for_each_descendant_pre(@pos, @css) {
842 * Update @css's state;
844 * Verify @pos is alive and inherit state from its parent;
849 * As long as the inheriting step, including checking the parent state, is
850 * enclosed inside @pos locking, double-locking the parent isn't necessary
851 * while inheriting. The state update to the parent is guaranteed to be
852 * visible by walking order and, as long as inheriting operations to the
853 * same @pos are atomic to each other, multiple updates racing each other
854 * still result in the correct state. It's guaranateed that at least one
855 * inheritance happens for any css after the latest update to its parent.
857 * If checking parent's state requires locking the parent, each inheriting
858 * iteration should lock and unlock both @pos->parent and @pos.
860 * Alternatively, a subsystem may choose to use a single global lock to
861 * synchronize ->css_online() and ->css_offline() against tree-walking
864 * It is allowed to temporarily drop RCU read lock during iteration. The
865 * caller is responsible for ensuring that @pos remains accessible until
866 * the start of the next iteration by, for example, bumping the css refcnt.
868 #define css_for_each_descendant_pre(pos, css) \
869 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
870 (pos) = css_next_descendant_pre((pos), (css)))
872 struct cgroup_subsys_state *
873 css_next_descendant_post(struct cgroup_subsys_state *pos,
874 struct cgroup_subsys_state *css);
877 * css_for_each_descendant_post - post-order walk of a css's descendants
878 * @pos: the css * to use as the loop cursor
879 * @css: css whose descendants to walk
881 * Similar to css_for_each_descendant_pre() but performs post-order
882 * traversal instead. @root is included in the iteration and the last
883 * node to be visited. Note that the walk visibility guarantee described
884 * in pre-order walk doesn't apply the same to post-order walks.
886 #define css_for_each_descendant_post(pos, css) \
887 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
888 (pos) = css_next_descendant_post((pos), (css)))
890 /* A css_task_iter should be treated as an opaque object */
891 struct css_task_iter {
892 struct cgroup_subsys *ss;
894 struct list_head *cset_pos;
895 struct list_head *cset_head;
897 struct list_head *task_pos;
898 struct list_head *tasks_head;
899 struct list_head *mg_tasks_head;
902 void css_task_iter_start(struct cgroup_subsys_state *css,
903 struct css_task_iter *it);
904 struct task_struct *css_task_iter_next(struct css_task_iter *it);
905 void css_task_iter_end(struct css_task_iter *it);
907 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
908 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
910 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
911 struct cgroup_subsys *ss);
913 #else /* !CONFIG_CGROUPS */
915 static inline int cgroup_init_early(void) { return 0; }
916 static inline int cgroup_init(void) { return 0; }
917 static inline void cgroup_fork(struct task_struct *p) {}
918 static inline void cgroup_post_fork(struct task_struct *p) {}
919 static inline void cgroup_exit(struct task_struct *p) {}
921 static inline int cgroupstats_build(struct cgroupstats *stats,
922 struct dentry *dentry)
927 /* No cgroups - nothing to do */
928 static inline int cgroup_attach_task_all(struct task_struct *from,
929 struct task_struct *t)
934 #endif /* !CONFIG_CGROUPS */
936 #endif /* _LINUX_CGROUP_H */