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/prio_heap.h>
18 #include <linux/rwsem.h>
19 #include <linux/idr.h>
20 #include <linux/workqueue.h>
21 #include <linux/xattr.h>
23 #include <linux/percpu-refcount.h>
34 extern int cgroup_init_early(void);
35 extern int cgroup_init(void);
36 extern void cgroup_fork(struct task_struct *p);
37 extern void cgroup_post_fork(struct task_struct *p);
38 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
39 extern int cgroupstats_build(struct cgroupstats *stats,
40 struct dentry *dentry);
41 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
42 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
44 extern int proc_cgroup_show(struct seq_file *, void *);
47 * Define the enumeration of all cgroup subsystems.
49 * We define ids for builtin subsystems and then modular ones.
51 #define SUBSYS(_x) _x ## _subsys_id,
52 enum cgroup_subsys_id {
53 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
54 #include <linux/cgroup_subsys.h>
55 #undef IS_SUBSYS_ENABLED
56 CGROUP_BUILTIN_SUBSYS_COUNT,
58 __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
60 #define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
61 #include <linux/cgroup_subsys.h>
62 #undef IS_SUBSYS_ENABLED
67 /* Per-subsystem/per-cgroup state maintained by the system. */
68 struct cgroup_subsys_state {
69 /* the cgroup that this css is attached to */
70 struct cgroup *cgroup;
72 /* the cgroup subsystem that this css is attached to */
73 struct cgroup_subsys *ss;
75 /* reference count - access via css_[try]get() and css_put() */
76 struct percpu_ref refcnt;
79 struct cgroup_subsys_state *parent;
82 /* ID for this css, if possible */
83 struct css_id __rcu *id;
85 /* Used to put @cgroup->dentry on the last css_put() */
86 struct work_struct destroy_work;
89 /* bits in struct cgroup_subsys_state flags field */
91 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
92 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
96 * css_get - obtain a reference on the specified css
99 * The caller must already have a reference.
101 static inline void css_get(struct cgroup_subsys_state *css)
103 /* We don't need to reference count the root state */
104 if (!(css->flags & CSS_ROOT))
105 percpu_ref_get(&css->refcnt);
109 * css_tryget - try to obtain a reference on the specified css
112 * Obtain a reference on @css if it's alive. The caller naturally needs to
113 * ensure that @css is accessible but doesn't have to be holding a
114 * reference on it - IOW, RCU protected access is good enough for this
115 * function. Returns %true if a reference count was successfully obtained;
118 static inline bool css_tryget(struct cgroup_subsys_state *css)
120 if (css->flags & CSS_ROOT)
122 return percpu_ref_tryget(&css->refcnt);
126 * css_put - put a css reference
129 * Put a reference obtained via css_get() and css_tryget().
131 static inline void css_put(struct cgroup_subsys_state *css)
133 if (!(css->flags & CSS_ROOT))
134 percpu_ref_put(&css->refcnt);
137 /* bits in struct cgroup flags field */
139 /* Control Group is dead */
142 * Control Group has previously had a child cgroup or a task,
143 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
146 /* Control Group requires release notifications to userspace */
147 CGRP_NOTIFY_ON_RELEASE,
149 * Clone the parent's configuration when creating a new child
150 * cpuset cgroup. For historical reasons, this option can be
151 * specified at mount time and thus is implemented here.
153 CGRP_CPUSET_CLONE_CHILDREN,
154 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
159 struct rcu_head rcu_head;
164 unsigned long flags; /* "unsigned long" so bitops work */
167 * idr allocated in-hierarchy ID.
169 * The ID of the root cgroup is always 0, and a new cgroup
170 * will be assigned with a smallest available ID.
175 * We link our 'sibling' struct into our parent's 'children'.
176 * Our children link their 'sibling' into our 'children'.
178 struct list_head sibling; /* my parent's children */
179 struct list_head children; /* my children */
180 struct list_head files; /* my files */
182 struct cgroup *parent; /* my parent */
183 struct dentry *dentry; /* cgroup fs entry, RCU protected */
186 * Monotonically increasing unique serial number which defines a
187 * uniform order among all cgroups. It's guaranteed that all
188 * ->children lists are in the ascending order of ->serial_nr.
189 * It's used to allow interrupting and resuming iterations.
194 * This is a copy of dentry->d_name, and it's needed because
195 * we can't use dentry->d_name in cgroup_path().
197 * You must acquire rcu_read_lock() to access cgrp->name, and
198 * the only place that can change it is rename(), which is
199 * protected by parent dir's i_mutex.
201 * Normally you should use cgroup_name() wrapper rather than
202 * access it directly.
204 struct cgroup_name __rcu *name;
206 /* Private pointers for each registered subsystem */
207 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
209 struct cgroupfs_root *root;
212 * List of cgrp_cset_links pointing at css_sets with tasks in this
213 * cgroup. Protected by css_set_lock.
215 struct list_head cset_links;
218 * Linked list running through all cgroups that can
219 * potentially be reaped by the release agent. Protected by
222 struct list_head release_list;
225 * list of pidlists, up to two for each namespace (one for procs, one
226 * for tasks); created on demand.
228 struct list_head pidlists;
229 struct mutex pidlist_mutex;
231 /* dummy css with NULL ->ss, points back to this cgroup */
232 struct cgroup_subsys_state dummy_css;
234 /* For css percpu_ref killing and RCU-protected deletion */
235 struct rcu_head rcu_head;
236 struct work_struct destroy_work;
237 atomic_t css_kill_cnt;
239 /* List of events which userspace want to receive */
240 struct list_head event_list;
241 spinlock_t event_list_lock;
243 /* directory xattrs */
244 struct simple_xattrs xattrs;
247 #define MAX_CGROUP_ROOT_NAMELEN 64
249 /* cgroupfs_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" and "clone_children" are disallowed.
268 * Also, cgroupfs file cgroup.clone_children is not created.
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 * - "release_agent" and "notify_on_release" are removed.
281 * Replacement notification mechanism will be implemented.
283 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
284 * and take masks of ancestors with non-empty cpus/mems, instead of
285 * being moved to an ancestor.
287 * - cpuset: a task can be moved into an empty cpuset, and again it
288 * takes masks of ancestors.
290 * - memcg: use_hierarchy is on by default and the cgroup file for
291 * the flag is not created.
293 * - blkcg: blk-throttle becomes properly hierarchical.
295 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
297 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
298 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
300 /* mount options live below bit 16 */
301 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
303 CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */
307 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
308 * associated with a superblock to form an active hierarchy. This is
309 * internal to cgroup core. Don't access directly from controllers.
311 struct cgroupfs_root {
312 struct super_block *sb;
314 /* The bitmask of subsystems attached to this hierarchy */
315 unsigned long subsys_mask;
317 /* Unique id for this hierarchy. */
320 /* A list running through the attached subsystems */
321 struct list_head subsys_list;
323 /* The root cgroup for this hierarchy */
324 struct cgroup top_cgroup;
326 /* Tracks how many cgroups are currently defined in hierarchy.*/
327 int number_of_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 * List running through all tasks using this cgroup
366 * group. Protected by css_set_lock
368 struct list_head tasks;
371 * List of cgrp_cset_links pointing at cgroups referenced from this
372 * css_set. Protected by css_set_lock.
374 struct list_head cgrp_links;
377 * Set of subsystem states, one for each subsystem. This array
378 * is immutable after creation apart from the init_css_set
379 * during subsystem registration (at boot time) and modular subsystem
382 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
384 /* For RCU-protected deletion */
385 struct rcu_head rcu_head;
389 * cgroup_map_cb is an abstract callback API for reporting map-valued
393 struct cgroup_map_cb {
394 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
399 * struct cftype: handler definitions for cgroup control files
401 * When reading/writing to a file:
402 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
403 * - the 'cftype' of the file is file->f_dentry->d_fsdata
408 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
409 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
410 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
413 #define MAX_CFTYPE_NAME 64
417 * By convention, the name should begin with the name of the
418 * subsystem, followed by a period. Zero length string indicates
419 * end of cftype array.
421 char name[MAX_CFTYPE_NAME];
424 * If not 0, file mode is set to this value, otherwise it will
425 * be figured out automatically
430 * If non-zero, defines the maximum length of string that can
431 * be passed to write_string; defaults to 64
433 size_t max_write_len;
439 * The subsys this file belongs to. Initialized automatically
440 * during registration. NULL for cgroup core files.
442 struct cgroup_subsys *ss;
444 int (*open)(struct inode *inode, struct file *file);
445 ssize_t (*read)(struct cgroup_subsys_state *css, struct cftype *cft,
447 char __user *buf, size_t nbytes, loff_t *ppos);
449 * read_u64() is a shortcut for the common case of returning a
450 * single integer. Use it in place of read()
452 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
454 * read_s64() is a signed version of read_u64()
456 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
458 * read_map() is used for defining a map of key/value
459 * pairs. It should call cb->fill(cb, key, value) for each
460 * entry. The key/value pairs (and their ordering) should not
461 * change between reboots.
463 int (*read_map)(struct cgroup_subsys_state *css, struct cftype *cft,
464 struct cgroup_map_cb *cb);
466 * read_seq_string() is used for outputting a simple sequence
469 int (*read_seq_string)(struct cgroup_subsys_state *css,
470 struct cftype *cft, struct seq_file *m);
472 ssize_t (*write)(struct cgroup_subsys_state *css, struct cftype *cft,
474 const char __user *buf, size_t nbytes, loff_t *ppos);
477 * write_u64() is a shortcut for the common case of accepting
478 * a single integer (as parsed by simple_strtoull) from
479 * userspace. Use in place of write(); return 0 or error.
481 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
484 * write_s64() is a signed version of write_u64()
486 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
490 * write_string() is passed a nul-terminated kernelspace
491 * buffer of maximum length determined by max_write_len.
492 * Returns 0 or -ve error code.
494 int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft,
497 * trigger() callback can be used to get some kick from the
498 * userspace, when the actual string written is not important
499 * at all. The private field can be used to determine the
500 * kick type for multiplexing.
502 int (*trigger)(struct cgroup_subsys_state *css, unsigned int event);
504 int (*release)(struct inode *inode, struct file *file);
507 * register_event() callback will be used to add new userspace
508 * waiter for changes related to the cftype. Implement it if
509 * you want to provide this functionality. Use eventfd_signal()
510 * on eventfd to send notification to userspace.
512 int (*register_event)(struct cgroup_subsys_state *css,
513 struct cftype *cft, struct eventfd_ctx *eventfd,
516 * unregister_event() callback will be called when userspace
517 * closes the eventfd or on cgroup removing.
518 * This callback must be implemented, if you want provide
519 * notification functionality.
521 void (*unregister_event)(struct cgroup_subsys_state *css,
523 struct eventfd_ctx *eventfd);
527 * cftype_sets describe cftypes belonging to a subsystem and are chained at
528 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
529 * terminated by zero length name.
532 struct list_head node; /* chained at subsys->cftsets */
537 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
538 * function can be called as long as @cgrp is accessible.
540 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
542 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
545 /* Caller should hold rcu_read_lock() */
546 static inline const char *cgroup_name(const struct cgroup *cgrp)
548 return rcu_dereference(cgrp->name)->name;
551 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
552 int cgroup_rm_cftypes(struct cftype *cfts);
554 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
556 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
557 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
559 int cgroup_task_count(const struct cgroup *cgrp);
562 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
565 struct cgroup_taskset;
566 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
567 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
568 struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset,
570 int cgroup_taskset_size(struct cgroup_taskset *tset);
573 * cgroup_taskset_for_each - iterate cgroup_taskset
574 * @task: the loop cursor
575 * @skip_css: skip if task's css matches this, %NULL to iterate through all
576 * @tset: taskset to iterate
578 #define cgroup_taskset_for_each(task, skip_css, tset) \
579 for ((task) = cgroup_taskset_first((tset)); (task); \
580 (task) = cgroup_taskset_next((tset))) \
582 cgroup_taskset_cur_css((tset), \
583 (skip_css)->ss->subsys_id) != (skip_css))
586 * Control Group subsystem type.
587 * See Documentation/cgroups/cgroups.txt for details
590 struct cgroup_subsys {
591 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
592 int (*css_online)(struct cgroup_subsys_state *css);
593 void (*css_offline)(struct cgroup_subsys_state *css);
594 void (*css_free)(struct cgroup_subsys_state *css);
596 int (*can_attach)(struct cgroup_subsys_state *css,
597 struct cgroup_taskset *tset);
598 void (*cancel_attach)(struct cgroup_subsys_state *css,
599 struct cgroup_taskset *tset);
600 void (*attach)(struct cgroup_subsys_state *css,
601 struct cgroup_taskset *tset);
602 void (*fork)(struct task_struct *task);
603 void (*exit)(struct cgroup_subsys_state *css,
604 struct cgroup_subsys_state *old_css,
605 struct task_struct *task);
606 void (*bind)(struct cgroup_subsys_state *root_css);
612 * True if this subsys uses ID. ID is not available before cgroup_init()
613 * (not available in early_init time.)
618 * If %false, this subsystem is properly hierarchical -
619 * configuration, resource accounting and restriction on a parent
620 * cgroup cover those of its children. If %true, hierarchy support
621 * is broken in some ways - some subsystems ignore hierarchy
622 * completely while others are only implemented half-way.
624 * It's now disallowed to create nested cgroups if the subsystem is
625 * broken and cgroup core will emit a warning message on such
626 * cases. Eventually, all subsystems will be made properly
627 * hierarchical and this will go away.
629 bool broken_hierarchy;
630 bool warned_broken_hierarchy;
632 #define MAX_CGROUP_TYPE_NAMELEN 32
636 * Link to parent, and list entry in parent's children.
637 * Protected by cgroup_lock()
639 struct cgroupfs_root *root;
640 struct list_head sibling;
641 /* used when use_id == true */
645 /* list of cftype_sets */
646 struct list_head cftsets;
648 /* base cftypes, automatically [de]registered with subsys itself */
649 struct cftype *base_cftypes;
650 struct cftype_set base_cftset;
652 /* should be defined only by modular subsystems */
653 struct module *module;
656 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
657 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
658 #include <linux/cgroup_subsys.h>
659 #undef IS_SUBSYS_ENABLED
663 * css_parent - find the parent css
664 * @css: the target cgroup_subsys_state
666 * Return the parent css of @css. This function is guaranteed to return
667 * non-NULL parent as long as @css isn't the root.
670 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
676 * task_css_set_check - obtain a task's css_set with extra access conditions
677 * @task: the task to obtain css_set for
678 * @__c: extra condition expression to be passed to rcu_dereference_check()
680 * A task's css_set is RCU protected, initialized and exited while holding
681 * task_lock(), and can only be modified while holding both cgroup_mutex
682 * and task_lock() while the task is alive. This macro verifies that the
683 * caller is inside proper critical section and returns @task's css_set.
685 * The caller can also specify additional allowed conditions via @__c, such
686 * as locks used during the cgroup_subsys::attach() methods.
688 #ifdef CONFIG_PROVE_RCU
689 extern struct mutex cgroup_mutex;
690 #define task_css_set_check(task, __c) \
691 rcu_dereference_check((task)->cgroups, \
692 lockdep_is_held(&(task)->alloc_lock) || \
693 lockdep_is_held(&cgroup_mutex) || (__c))
695 #define task_css_set_check(task, __c) \
696 rcu_dereference((task)->cgroups)
700 * task_css_check - obtain css for (task, subsys) w/ extra access conds
701 * @task: the target task
702 * @subsys_id: the target subsystem ID
703 * @__c: extra condition expression to be passed to rcu_dereference_check()
705 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
706 * synchronization rules are the same as task_css_set_check().
708 #define task_css_check(task, subsys_id, __c) \
709 task_css_set_check((task), (__c))->subsys[(subsys_id)]
712 * task_css_set - obtain a task's css_set
713 * @task: the task to obtain css_set for
715 * See task_css_set_check().
717 static inline struct css_set *task_css_set(struct task_struct *task)
719 return task_css_set_check(task, false);
723 * task_css - obtain css for (task, subsys)
724 * @task: the target task
725 * @subsys_id: the target subsystem ID
727 * See task_css_check().
729 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
732 return task_css_check(task, subsys_id, false);
735 static inline struct cgroup *task_cgroup(struct task_struct *task,
738 return task_css(task, subsys_id)->cgroup;
742 * cgroup_from_id - lookup cgroup by id
743 * @ss: cgroup subsys to be looked into
746 * Returns the cgroup if there's valid one with @id, otherwise returns NULL.
747 * Should be called under rcu_read_lock().
749 static inline struct cgroup *cgroup_from_id(struct cgroup_subsys *ss, int id)
751 #ifdef CONFIG_PROVE_RCU
752 rcu_lockdep_assert(rcu_read_lock_held() ||
753 lockdep_is_held(&cgroup_mutex),
754 "cgroup_from_id() needs proper protection");
756 return idr_find(&ss->root->cgroup_idr, id);
759 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
760 struct cgroup_subsys_state *parent);
763 * css_for_each_child - iterate through children of a css
764 * @pos: the css * to use as the loop cursor
765 * @parent: css whose children to walk
767 * Walk @parent's children. Must be called under rcu_read_lock(). A child
768 * css which hasn't finished ->css_online() or already has finished
769 * ->css_offline() may show up during traversal and it's each subsystem's
770 * responsibility to verify that each @pos is alive.
772 * If a subsystem synchronizes against the parent in its ->css_online() and
773 * before starting iterating, a css which finished ->css_online() is
774 * guaranteed to be visible in the future iterations.
776 * It is allowed to temporarily drop RCU read lock during iteration. The
777 * caller is responsible for ensuring that @pos remains accessible until
778 * the start of the next iteration by, for example, bumping the css refcnt.
780 #define css_for_each_child(pos, parent) \
781 for ((pos) = css_next_child(NULL, (parent)); (pos); \
782 (pos) = css_next_child((pos), (parent)))
784 struct cgroup_subsys_state *
785 css_next_descendant_pre(struct cgroup_subsys_state *pos,
786 struct cgroup_subsys_state *css);
788 struct cgroup_subsys_state *
789 css_rightmost_descendant(struct cgroup_subsys_state *pos);
792 * css_for_each_descendant_pre - pre-order walk of a css's descendants
793 * @pos: the css * to use as the loop cursor
794 * @root: css whose descendants to walk
796 * Walk @root's descendants. @root is included in the iteration and the
797 * first node to be visited. Must be called under rcu_read_lock(). A
798 * descendant css which hasn't finished ->css_online() or already has
799 * finished ->css_offline() may show up during traversal and it's each
800 * subsystem's responsibility to verify that each @pos is alive.
802 * If a subsystem synchronizes against the parent in its ->css_online() and
803 * before starting iterating, and synchronizes against @pos on each
804 * iteration, any descendant css which finished ->css_online() is
805 * guaranteed to be visible in the future iterations.
807 * In other words, the following guarantees that a descendant can't escape
808 * state updates of its ancestors.
812 * Lock @css's parent and @css;
813 * Inherit state from the parent;
817 * my_update_state(@css)
819 * css_for_each_descendant_pre(@pos, @css) {
822 * Update @css's state;
824 * Verify @pos is alive and inherit state from its parent;
829 * As long as the inheriting step, including checking the parent state, is
830 * enclosed inside @pos locking, double-locking the parent isn't necessary
831 * while inheriting. The state update to the parent is guaranteed to be
832 * visible by walking order and, as long as inheriting operations to the
833 * same @pos are atomic to each other, multiple updates racing each other
834 * still result in the correct state. It's guaranateed that at least one
835 * inheritance happens for any css after the latest update to its parent.
837 * If checking parent's state requires locking the parent, each inheriting
838 * iteration should lock and unlock both @pos->parent and @pos.
840 * Alternatively, a subsystem may choose to use a single global lock to
841 * synchronize ->css_online() and ->css_offline() against tree-walking
844 * It is allowed to temporarily drop RCU read lock during iteration. The
845 * caller is responsible for ensuring that @pos remains accessible until
846 * the start of the next iteration by, for example, bumping the css refcnt.
848 #define css_for_each_descendant_pre(pos, css) \
849 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
850 (pos) = css_next_descendant_pre((pos), (css)))
852 struct cgroup_subsys_state *
853 css_next_descendant_post(struct cgroup_subsys_state *pos,
854 struct cgroup_subsys_state *css);
857 * css_for_each_descendant_post - post-order walk of a css's descendants
858 * @pos: the css * to use as the loop cursor
859 * @css: css whose descendants to walk
861 * Similar to css_for_each_descendant_pre() but performs post-order
862 * traversal instead. @root is included in the iteration and the last
863 * node to be visited. Note that the walk visibility guarantee described
864 * in pre-order walk doesn't apply the same to post-order walks.
866 #define css_for_each_descendant_post(pos, css) \
867 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
868 (pos) = css_next_descendant_post((pos), (css)))
870 /* A css_task_iter should be treated as an opaque object */
871 struct css_task_iter {
872 struct cgroup_subsys_state *origin_css;
873 struct list_head *cset_link;
874 struct list_head *task;
877 void css_task_iter_start(struct cgroup_subsys_state *css,
878 struct css_task_iter *it);
879 struct task_struct *css_task_iter_next(struct css_task_iter *it);
880 void css_task_iter_end(struct css_task_iter *it);
882 int css_scan_tasks(struct cgroup_subsys_state *css,
883 bool (*test)(struct task_struct *, void *),
884 void (*process)(struct task_struct *, void *),
885 void *data, struct ptr_heap *heap);
887 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
888 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
891 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
892 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
893 * CSS ID is assigned at cgroup allocation (create) automatically
894 * and removed when subsys calls free_css_id() function. This is because
895 * the lifetime of cgroup_subsys_state is subsys's matter.
897 * Looking up and scanning function should be called under rcu_read_lock().
898 * Taking cgroup_mutex is not necessary for following calls.
899 * But the css returned by this routine can be "not populated yet" or "being
900 * destroyed". The caller should check css and cgroup's status.
904 * Typically Called at ->destroy(), or somewhere the subsys frees
905 * cgroup_subsys_state.
907 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
909 /* Find a cgroup_subsys_state which has given ID */
911 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
913 /* Returns true if root is ancestor of cg */
914 bool css_is_ancestor(struct cgroup_subsys_state *cg,
915 const struct cgroup_subsys_state *root);
917 /* Get id and depth of css */
918 unsigned short css_id(struct cgroup_subsys_state *css);
919 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
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, int callbacks) {}
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 */