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 *control_kn; /* kn for "cgroup.subtree_control" */
181 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
184 * Monotonically increasing unique serial number which defines a
185 * uniform order among all cgroups. It's guaranteed that all
186 * ->children lists are in the ascending order of ->serial_nr.
187 * It's used to allow interrupting and resuming iterations.
191 /* the bitmask of subsystems enabled on the child cgroups */
192 unsigned int child_subsys_mask;
194 /* Private pointers for each registered subsystem */
195 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
197 struct cgroup_root *root;
200 * List of cgrp_cset_links pointing at css_sets with tasks in this
201 * cgroup. Protected by css_set_lock.
203 struct list_head cset_links;
206 * On the default hierarchy, a css_set for a cgroup with some
207 * susbsys disabled will point to css's which are associated with
208 * the closest ancestor which has the subsys enabled. The
209 * following lists all css_sets which point to this cgroup's css
210 * for the given subsystem.
212 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
215 * Linked list running through all cgroups that can
216 * potentially be reaped by the release agent. Protected by
219 struct list_head release_list;
222 * list of pidlists, up to two for each namespace (one for procs, one
223 * for tasks); created on demand.
225 struct list_head pidlists;
226 struct mutex pidlist_mutex;
228 /* dummy css with NULL ->ss, points back to this cgroup */
229 struct cgroup_subsys_state dummy_css;
231 /* For css percpu_ref killing and RCU-protected deletion */
232 struct rcu_head rcu_head;
233 struct work_struct destroy_work;
235 /* used to wait for offlining of csses */
236 wait_queue_head_t offline_waitq;
239 #define MAX_CGROUP_ROOT_NAMELEN 64
241 /* cgroup_root->flags */
244 * Unfortunately, cgroup core and various controllers are riddled
245 * with idiosyncrasies and pointless options. The following flag,
246 * when set, will force sane behavior - some options are forced on,
247 * others are disallowed, and some controllers will change their
248 * hierarchical or other behaviors.
250 * The set of behaviors affected by this flag are still being
251 * determined and developed and the mount option for this flag is
252 * prefixed with __DEVEL__. The prefix will be dropped once we
253 * reach the point where all behaviors are compatible with the
254 * planned unified hierarchy, which will automatically turn on this
257 * The followings are the behaviors currently affected this flag.
259 * - Mount options "noprefix", "xattr", "clone_children",
260 * "release_agent" and "name" are disallowed.
262 * - When mounting an existing superblock, mount options should
265 * - Remount is disallowed.
267 * - rename(2) is disallowed.
269 * - "tasks" is removed. Everything should be at process
270 * granularity. Use "cgroup.procs" instead.
272 * - "cgroup.procs" is not sorted. pids will be unique unless they
273 * got recycled inbetween reads.
275 * - "release_agent" and "notify_on_release" are removed.
276 * Replacement notification mechanism will be implemented.
278 * - "cgroup.clone_children" is removed.
280 * - "cgroup.subtree_populated" is available. Its value is 0 if
281 * the cgroup and its descendants contain no task; otherwise, 1.
282 * The file also generates kernfs notification which can be
283 * monitored through poll and [di]notify when the value of the
286 * - If mount is requested with sane_behavior but without any
287 * subsystem, the default unified hierarchy is mounted.
289 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
290 * and take masks of ancestors with non-empty cpus/mems, instead of
291 * being moved to an ancestor.
293 * - cpuset: a task can be moved into an empty cpuset, and again it
294 * takes masks of ancestors.
296 * - memcg: use_hierarchy is on by default and the cgroup file for
297 * the flag is not created.
299 * - blkcg: blk-throttle becomes properly hierarchical.
301 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
303 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
304 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
306 /* mount options live below bit 16 */
307 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
311 * A cgroup_root represents the root of a cgroup hierarchy, and may be
312 * associated with a kernfs_root to form an active hierarchy. This is
313 * internal to cgroup core. Don't access directly from controllers.
316 struct kernfs_root *kf_root;
318 /* The bitmask of subsystems attached to this hierarchy */
319 unsigned int subsys_mask;
321 /* Unique id for this hierarchy. */
324 /* The root cgroup. Root is destroyed on its release. */
327 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
330 /* A list running through the active hierarchies */
331 struct list_head root_list;
333 /* Hierarchy-specific flags */
336 /* IDs for cgroups in this hierarchy */
337 struct idr cgroup_idr;
339 /* The path to use for release notifications. */
340 char release_agent_path[PATH_MAX];
342 /* The name for this hierarchy - may be empty */
343 char name[MAX_CGROUP_ROOT_NAMELEN];
347 * A css_set is a structure holding pointers to a set of
348 * cgroup_subsys_state objects. This saves space in the task struct
349 * object and speeds up fork()/exit(), since a single inc/dec and a
350 * list_add()/del() can bump the reference count on the entire cgroup
356 /* Reference count */
360 * List running through all cgroup groups in the same hash
361 * slot. Protected by css_set_lock
363 struct hlist_node hlist;
366 * Lists running through all tasks using this cgroup group.
367 * mg_tasks lists tasks which belong to this cset but are in the
368 * process of being migrated out or in. Protected by
369 * css_set_rwsem, but, during migration, once tasks are moved to
370 * mg_tasks, it can be read safely while holding cgroup_mutex.
372 struct list_head tasks;
373 struct list_head mg_tasks;
376 * List of cgrp_cset_links pointing at cgroups referenced from this
377 * css_set. Protected by css_set_lock.
379 struct list_head cgrp_links;
381 /* the default cgroup associated with this css_set */
382 struct cgroup *dfl_cgrp;
385 * Set of subsystem states, one for each subsystem. This array is
386 * immutable after creation apart from the init_css_set during
387 * subsystem registration (at boot time).
389 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
392 * List of csets participating in the on-going migration either as
393 * source or destination. Protected by cgroup_mutex.
395 struct list_head mg_preload_node;
396 struct list_head mg_node;
399 * If this cset is acting as the source of migration the following
400 * two fields are set. mg_src_cgrp is the source cgroup of the
401 * on-going migration and mg_dst_cset is the destination cset the
402 * target tasks on this cset should be migrated to. Protected by
405 struct cgroup *mg_src_cgrp;
406 struct css_set *mg_dst_cset;
409 * On the default hierarhcy, ->subsys[ssid] may point to a css
410 * attached to an ancestor instead of the cgroup this css_set is
411 * associated with. The following node is anchored at
412 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
413 * iterate through all css's attached to a given cgroup.
415 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
417 /* For RCU-protected deletion */
418 struct rcu_head rcu_head;
422 * struct cftype: handler definitions for cgroup control files
424 * When reading/writing to a file:
425 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
426 * - the 'cftype' of the file is file->f_dentry->d_fsdata
431 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
432 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
433 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
434 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
435 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
438 #define MAX_CFTYPE_NAME 64
442 * By convention, the name should begin with the name of the
443 * subsystem, followed by a period. Zero length string indicates
444 * end of cftype array.
446 char name[MAX_CFTYPE_NAME];
449 * If not 0, file mode is set to this value, otherwise it will
450 * be figured out automatically
455 * The maximum length of string, excluding trailing nul, that can
456 * be passed to write_string. If < PAGE_SIZE-1, PAGE_SIZE-1 is
459 size_t max_write_len;
465 * Fields used for internal bookkeeping. Initialized automatically
466 * during registration.
468 struct cgroup_subsys *ss; /* NULL for cgroup core files */
469 struct list_head node; /* anchored at ss->cfts */
470 struct kernfs_ops *kf_ops;
473 * read_u64() is a shortcut for the common case of returning a
474 * single integer. Use it in place of read()
476 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
478 * read_s64() is a signed version of read_u64()
480 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
482 /* generic seq_file read interface */
483 int (*seq_show)(struct seq_file *sf, void *v);
485 /* optional ops, implement all or none */
486 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
487 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
488 void (*seq_stop)(struct seq_file *sf, void *v);
491 * write_u64() is a shortcut for the common case of accepting
492 * a single integer (as parsed by simple_strtoull) from
493 * userspace. Use in place of write(); return 0 or error.
495 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
498 * write_s64() is a signed version of write_u64()
500 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
504 * write_string() is passed a nul-terminated kernelspace
505 * buffer of maximum length determined by max_write_len.
506 * Returns 0 or -ve error code.
508 int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft,
511 * trigger() callback can be used to get some kick from the
512 * userspace, when the actual string written is not important
513 * at all. The private field can be used to determine the
514 * kick type for multiplexing.
516 int (*trigger)(struct cgroup_subsys_state *css, unsigned int event);
518 #ifdef CONFIG_DEBUG_LOCK_ALLOC
519 struct lock_class_key lockdep_key;
523 extern struct cgroup_root cgrp_dfl_root;
524 extern struct css_set init_css_set;
526 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
528 return cgrp->root == &cgrp_dfl_root;
532 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
533 * function can be called as long as @cgrp is accessible.
535 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
537 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
540 /* no synchronization, the result can only be used as a hint */
541 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
543 return !list_empty(&cgrp->cset_links);
546 /* returns ino associated with a cgroup, 0 indicates unmounted root */
547 static inline ino_t cgroup_ino(struct cgroup *cgrp)
550 return cgrp->kn->ino;
555 static inline struct cftype *seq_cft(struct seq_file *seq)
557 struct kernfs_open_file *of = seq->private;
562 struct cgroup_subsys_state *seq_css(struct seq_file *seq);
565 * Name / path handling functions. All are thin wrappers around the kernfs
566 * counterparts and can be called under any context.
569 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
571 return kernfs_name(cgrp->kn, buf, buflen);
574 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
577 return kernfs_path(cgrp->kn, buf, buflen);
580 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
582 pr_cont_kernfs_name(cgrp->kn);
585 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
587 pr_cont_kernfs_path(cgrp->kn);
590 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
592 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
593 int cgroup_rm_cftypes(struct cftype *cfts);
595 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
598 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
601 struct cgroup_taskset;
602 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
603 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
606 * cgroup_taskset_for_each - iterate cgroup_taskset
607 * @task: the loop cursor
608 * @tset: taskset to iterate
610 #define cgroup_taskset_for_each(task, tset) \
611 for ((task) = cgroup_taskset_first((tset)); (task); \
612 (task) = cgroup_taskset_next((tset)))
615 * Control Group subsystem type.
616 * See Documentation/cgroups/cgroups.txt for details
619 struct cgroup_subsys {
620 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
621 int (*css_online)(struct cgroup_subsys_state *css);
622 void (*css_offline)(struct cgroup_subsys_state *css);
623 void (*css_free)(struct cgroup_subsys_state *css);
625 int (*can_attach)(struct cgroup_subsys_state *css,
626 struct cgroup_taskset *tset);
627 void (*cancel_attach)(struct cgroup_subsys_state *css,
628 struct cgroup_taskset *tset);
629 void (*attach)(struct cgroup_subsys_state *css,
630 struct cgroup_taskset *tset);
631 void (*fork)(struct task_struct *task);
632 void (*exit)(struct cgroup_subsys_state *css,
633 struct cgroup_subsys_state *old_css,
634 struct task_struct *task);
635 void (*bind)(struct cgroup_subsys_state *root_css);
641 * If %false, this subsystem is properly hierarchical -
642 * configuration, resource accounting and restriction on a parent
643 * cgroup cover those of its children. If %true, hierarchy support
644 * is broken in some ways - some subsystems ignore hierarchy
645 * completely while others are only implemented half-way.
647 * It's now disallowed to create nested cgroups if the subsystem is
648 * broken and cgroup core will emit a warning message on such
649 * cases. Eventually, all subsystems will be made properly
650 * hierarchical and this will go away.
652 bool broken_hierarchy;
653 bool warned_broken_hierarchy;
655 /* the following two fields are initialized automtically during boot */
657 #define MAX_CGROUP_TYPE_NAMELEN 32
660 /* link to parent, protected by cgroup_lock() */
661 struct cgroup_root *root;
663 /* idr for css->id */
667 * List of cftypes. Each entry is the first entry of an array
668 * terminated by zero length name.
670 struct list_head cfts;
672 /* base cftypes, automatically registered with subsys itself */
673 struct cftype *base_cftypes;
676 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
677 #include <linux/cgroup_subsys.h>
681 * css_parent - find the parent css
682 * @css: the target cgroup_subsys_state
684 * Return the parent css of @css. This function is guaranteed to return
685 * non-NULL parent as long as @css isn't the root.
688 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
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(). A child
786 * css which hasn't finished ->css_online() or already has finished
787 * ->css_offline() may show up during traversal and it's each subsystem's
788 * responsibility to verify that each @pos is alive.
790 * If a subsystem synchronizes against the parent in its ->css_online() and
791 * before starting iterating, a css which finished ->css_online() is
792 * guaranteed to be visible in the future iterations.
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(). A
816 * descendant css which hasn't finished ->css_online() or already has
817 * finished ->css_offline() may show up during traversal and it's each
818 * subsystem's responsibility to verify that each @pos is alive.
820 * If a subsystem synchronizes against the parent in its ->css_online() and
821 * before starting iterating, and synchronizes against @pos on each
822 * iteration, any descendant css which finished ->css_online() is
823 * guaranteed to be visible in the future iterations.
825 * In other words, the following guarantees that a descendant can't escape
826 * state updates of its ancestors.
830 * Lock @css's parent and @css;
831 * Inherit state from the parent;
835 * my_update_state(@css)
837 * css_for_each_descendant_pre(@pos, @css) {
840 * Update @css's state;
842 * Verify @pos is alive and inherit state from its parent;
847 * As long as the inheriting step, including checking the parent state, is
848 * enclosed inside @pos locking, double-locking the parent isn't necessary
849 * while inheriting. The state update to the parent is guaranteed to be
850 * visible by walking order and, as long as inheriting operations to the
851 * same @pos are atomic to each other, multiple updates racing each other
852 * still result in the correct state. It's guaranateed that at least one
853 * inheritance happens for any css after the latest update to its parent.
855 * If checking parent's state requires locking the parent, each inheriting
856 * iteration should lock and unlock both @pos->parent and @pos.
858 * Alternatively, a subsystem may choose to use a single global lock to
859 * synchronize ->css_online() and ->css_offline() against tree-walking
862 * It is allowed to temporarily drop RCU read lock during iteration. The
863 * caller is responsible for ensuring that @pos remains accessible until
864 * the start of the next iteration by, for example, bumping the css refcnt.
866 #define css_for_each_descendant_pre(pos, css) \
867 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
868 (pos) = css_next_descendant_pre((pos), (css)))
870 struct cgroup_subsys_state *
871 css_next_descendant_post(struct cgroup_subsys_state *pos,
872 struct cgroup_subsys_state *css);
875 * css_for_each_descendant_post - post-order walk of a css's descendants
876 * @pos: the css * to use as the loop cursor
877 * @css: css whose descendants to walk
879 * Similar to css_for_each_descendant_pre() but performs post-order
880 * traversal instead. @root is included in the iteration and the last
881 * node to be visited. Note that the walk visibility guarantee described
882 * in pre-order walk doesn't apply the same to post-order walks.
884 #define css_for_each_descendant_post(pos, css) \
885 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
886 (pos) = css_next_descendant_post((pos), (css)))
888 /* A css_task_iter should be treated as an opaque object */
889 struct css_task_iter {
890 struct cgroup_subsys *ss;
892 struct list_head *cset_pos;
893 struct list_head *cset_head;
895 struct list_head *task_pos;
896 struct list_head *tasks_head;
897 struct list_head *mg_tasks_head;
900 void css_task_iter_start(struct cgroup_subsys_state *css,
901 struct css_task_iter *it);
902 struct task_struct *css_task_iter_next(struct css_task_iter *it);
903 void css_task_iter_end(struct css_task_iter *it);
905 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
906 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
908 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
909 struct cgroup_subsys *ss);
911 #else /* !CONFIG_CGROUPS */
913 static inline int cgroup_init_early(void) { return 0; }
914 static inline int cgroup_init(void) { return 0; }
915 static inline void cgroup_fork(struct task_struct *p) {}
916 static inline void cgroup_post_fork(struct task_struct *p) {}
917 static inline void cgroup_exit(struct task_struct *p) {}
919 static inline int cgroupstats_build(struct cgroupstats *stats,
920 struct dentry *dentry)
925 /* No cgroups - nothing to do */
926 static inline int cgroup_attach_task_all(struct task_struct *from,
927 struct task_struct *t)
932 #endif /* !CONFIG_CGROUPS */
934 #endif /* _LINUX_CGROUP_H */