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 /* ID for this css, if possible */
80 struct css_id __rcu *id;
82 /* Used to put @cgroup->dentry on the last css_put() */
83 struct work_struct dput_work;
86 /* bits in struct cgroup_subsys_state flags field */
88 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
89 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
93 * css_get - obtain a reference on the specified css
96 * The caller must already have a reference.
98 static inline void css_get(struct cgroup_subsys_state *css)
100 /* We don't need to reference count the root state */
101 if (!(css->flags & CSS_ROOT))
102 percpu_ref_get(&css->refcnt);
106 * css_tryget - try to obtain a reference on the specified css
109 * Obtain a reference on @css if it's alive. The caller naturally needs to
110 * ensure that @css is accessible but doesn't have to be holding a
111 * reference on it - IOW, RCU protected access is good enough for this
112 * function. Returns %true if a reference count was successfully obtained;
115 static inline bool css_tryget(struct cgroup_subsys_state *css)
117 if (css->flags & CSS_ROOT)
119 return percpu_ref_tryget(&css->refcnt);
123 * css_put - put a css reference
126 * Put a reference obtained via css_get() and css_tryget().
128 static inline void css_put(struct cgroup_subsys_state *css)
130 if (!(css->flags & CSS_ROOT))
131 percpu_ref_put(&css->refcnt);
134 /* bits in struct cgroup flags field */
136 /* Control Group is dead */
139 * Control Group has previously had a child cgroup or a task,
140 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
143 /* Control Group requires release notifications to userspace */
144 CGRP_NOTIFY_ON_RELEASE,
146 * Clone the parent's configuration when creating a new child
147 * cpuset cgroup. For historical reasons, this option can be
148 * specified at mount time and thus is implemented here.
150 CGRP_CPUSET_CLONE_CHILDREN,
151 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
156 struct rcu_head rcu_head;
161 unsigned long flags; /* "unsigned long" so bitops work */
164 * idr allocated in-hierarchy ID.
166 * The ID of the root cgroup is always 0, and a new cgroup
167 * will be assigned with a smallest available ID.
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 */
177 struct list_head files; /* my files */
179 struct cgroup *parent; /* my parent */
180 struct dentry *dentry; /* cgroup fs entry, RCU protected */
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.
191 * This is a copy of dentry->d_name, and it's needed because
192 * we can't use dentry->d_name in cgroup_path().
194 * You must acquire rcu_read_lock() to access cgrp->name, and
195 * the only place that can change it is rename(), which is
196 * protected by parent dir's i_mutex.
198 * Normally you should use cgroup_name() wrapper rather than
199 * access it directly.
201 struct cgroup_name __rcu *name;
203 /* Private pointers for each registered subsystem */
204 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
206 struct cgroupfs_root *root;
209 * List of cgrp_cset_links pointing at css_sets with tasks in this
210 * cgroup. Protected by css_set_lock.
212 struct list_head cset_links;
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;
234 atomic_t css_kill_cnt;
236 /* List of events which userspace want to receive */
237 struct list_head event_list;
238 spinlock_t event_list_lock;
240 /* directory xattrs */
241 struct simple_xattrs xattrs;
244 #define MAX_CGROUP_ROOT_NAMELEN 64
246 /* cgroupfs_root->flags */
249 * Unfortunately, cgroup core and various controllers are riddled
250 * with idiosyncrasies and pointless options. The following flag,
251 * when set, will force sane behavior - some options are forced on,
252 * others are disallowed, and some controllers will change their
253 * hierarchical or other behaviors.
255 * The set of behaviors affected by this flag are still being
256 * determined and developed and the mount option for this flag is
257 * prefixed with __DEVEL__. The prefix will be dropped once we
258 * reach the point where all behaviors are compatible with the
259 * planned unified hierarchy, which will automatically turn on this
262 * The followings are the behaviors currently affected this flag.
264 * - Mount options "noprefix" and "clone_children" are disallowed.
265 * Also, cgroupfs file cgroup.clone_children is not created.
267 * - When mounting an existing superblock, mount options should
270 * - Remount is disallowed.
272 * - rename(2) is disallowed.
274 * - "tasks" is removed. Everything should be at process
275 * granularity. Use "cgroup.procs" instead.
277 * - "release_agent" and "notify_on_release" are removed.
278 * Replacement notification mechanism will be implemented.
280 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
281 * and take masks of ancestors with non-empty cpus/mems, instead of
282 * being moved to an ancestor.
284 * - cpuset: a task can be moved into an empty cpuset, and again it
285 * takes masks of ancestors.
287 * - memcg: use_hierarchy is on by default and the cgroup file for
288 * the flag is not created.
290 * - blkcg: blk-throttle becomes properly hierarchical.
292 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
294 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
295 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
297 /* mount options live below bit 16 */
298 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
300 CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */
304 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
305 * associated with a superblock to form an active hierarchy. This is
306 * internal to cgroup core. Don't access directly from controllers.
308 struct cgroupfs_root {
309 struct super_block *sb;
311 /* The bitmask of subsystems attached to this hierarchy */
312 unsigned long subsys_mask;
314 /* Unique id for this hierarchy. */
317 /* A list running through the attached subsystems */
318 struct list_head subsys_list;
320 /* The root cgroup for this hierarchy */
321 struct cgroup top_cgroup;
323 /* Tracks how many cgroups are currently defined in hierarchy.*/
324 int number_of_cgroups;
326 /* A list running through the active hierarchies */
327 struct list_head root_list;
329 /* Hierarchy-specific flags */
332 /* IDs for cgroups in this hierarchy */
333 struct idr cgroup_idr;
335 /* The path to use for release notifications. */
336 char release_agent_path[PATH_MAX];
338 /* The name for this hierarchy - may be empty */
339 char name[MAX_CGROUP_ROOT_NAMELEN];
343 * A css_set is a structure holding pointers to a set of
344 * cgroup_subsys_state objects. This saves space in the task struct
345 * object and speeds up fork()/exit(), since a single inc/dec and a
346 * list_add()/del() can bump the reference count on the entire cgroup
352 /* Reference count */
356 * List running through all cgroup groups in the same hash
357 * slot. Protected by css_set_lock
359 struct hlist_node hlist;
362 * List running through all tasks using this cgroup
363 * group. Protected by css_set_lock
365 struct list_head tasks;
368 * List of cgrp_cset_links pointing at cgroups referenced from this
369 * css_set. Protected by css_set_lock.
371 struct list_head cgrp_links;
374 * Set of subsystem states, one for each subsystem. This array
375 * is immutable after creation apart from the init_css_set
376 * during subsystem registration (at boot time) and modular subsystem
379 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
381 /* For RCU-protected deletion */
382 struct rcu_head rcu_head;
386 * cgroup_map_cb is an abstract callback API for reporting map-valued
390 struct cgroup_map_cb {
391 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
396 * struct cftype: handler definitions for cgroup control files
398 * When reading/writing to a file:
399 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
400 * - the 'cftype' of the file is file->f_dentry->d_fsdata
405 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
406 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
407 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
410 #define MAX_CFTYPE_NAME 64
414 * By convention, the name should begin with the name of the
415 * subsystem, followed by a period. Zero length string indicates
416 * end of cftype array.
418 char name[MAX_CFTYPE_NAME];
421 * If not 0, file mode is set to this value, otherwise it will
422 * be figured out automatically
427 * If non-zero, defines the maximum length of string that can
428 * be passed to write_string; defaults to 64
430 size_t max_write_len;
436 * The subsys this file belongs to. Initialized automatically
437 * during registration. NULL for cgroup core files.
439 struct cgroup_subsys *ss;
441 int (*open)(struct inode *inode, struct file *file);
442 ssize_t (*read)(struct cgroup_subsys_state *css, struct cftype *cft,
444 char __user *buf, size_t nbytes, loff_t *ppos);
446 * read_u64() is a shortcut for the common case of returning a
447 * single integer. Use it in place of read()
449 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
451 * read_s64() is a signed version of read_u64()
453 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
455 * read_map() is used for defining a map of key/value
456 * pairs. It should call cb->fill(cb, key, value) for each
457 * entry. The key/value pairs (and their ordering) should not
458 * change between reboots.
460 int (*read_map)(struct cgroup_subsys_state *css, struct cftype *cft,
461 struct cgroup_map_cb *cb);
463 * read_seq_string() is used for outputting a simple sequence
466 int (*read_seq_string)(struct cgroup_subsys_state *css,
467 struct cftype *cft, struct seq_file *m);
469 ssize_t (*write)(struct cgroup_subsys_state *css, struct cftype *cft,
471 const char __user *buf, size_t nbytes, loff_t *ppos);
474 * write_u64() is a shortcut for the common case of accepting
475 * a single integer (as parsed by simple_strtoull) from
476 * userspace. Use in place of write(); return 0 or error.
478 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
481 * write_s64() is a signed version of write_u64()
483 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
487 * write_string() is passed a nul-terminated kernelspace
488 * buffer of maximum length determined by max_write_len.
489 * Returns 0 or -ve error code.
491 int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft,
494 * trigger() callback can be used to get some kick from the
495 * userspace, when the actual string written is not important
496 * at all. The private field can be used to determine the
497 * kick type for multiplexing.
499 int (*trigger)(struct cgroup_subsys_state *css, unsigned int event);
501 int (*release)(struct inode *inode, struct file *file);
504 * register_event() callback will be used to add new userspace
505 * waiter for changes related to the cftype. Implement it if
506 * you want to provide this functionality. Use eventfd_signal()
507 * on eventfd to send notification to userspace.
509 int (*register_event)(struct cgroup_subsys_state *css,
510 struct cftype *cft, struct eventfd_ctx *eventfd,
513 * unregister_event() callback will be called when userspace
514 * closes the eventfd or on cgroup removing.
515 * This callback must be implemented, if you want provide
516 * notification functionality.
518 void (*unregister_event)(struct cgroup_subsys_state *css,
520 struct eventfd_ctx *eventfd);
524 * cftype_sets describe cftypes belonging to a subsystem and are chained at
525 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
526 * terminated by zero length name.
529 struct list_head node; /* chained at subsys->cftsets */
534 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
535 * function can be called as long as @cgrp is accessible.
537 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
539 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
542 /* Caller should hold rcu_read_lock() */
543 static inline const char *cgroup_name(const struct cgroup *cgrp)
545 return rcu_dereference(cgrp->name)->name;
548 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
549 int cgroup_rm_cftypes(struct cftype *cfts);
551 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
553 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
554 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
556 int cgroup_task_count(const struct cgroup *cgrp);
559 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
562 struct cgroup_taskset;
563 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
564 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
565 struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset,
567 int cgroup_taskset_size(struct cgroup_taskset *tset);
570 * cgroup_taskset_for_each - iterate cgroup_taskset
571 * @task: the loop cursor
572 * @skip_css: skip if task's css matches this, %NULL to iterate through all
573 * @tset: taskset to iterate
575 #define cgroup_taskset_for_each(task, skip_css, tset) \
576 for ((task) = cgroup_taskset_first((tset)); (task); \
577 (task) = cgroup_taskset_next((tset))) \
579 cgroup_taskset_cur_css((tset), \
580 (skip_css)->ss->subsys_id) != (skip_css))
583 * Control Group subsystem type.
584 * See Documentation/cgroups/cgroups.txt for details
587 struct cgroup_subsys {
588 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
589 int (*css_online)(struct cgroup_subsys_state *css);
590 void (*css_offline)(struct cgroup_subsys_state *css);
591 void (*css_free)(struct cgroup_subsys_state *css);
593 int (*can_attach)(struct cgroup_subsys_state *css,
594 struct cgroup_taskset *tset);
595 void (*cancel_attach)(struct cgroup_subsys_state *css,
596 struct cgroup_taskset *tset);
597 void (*attach)(struct cgroup_subsys_state *css,
598 struct cgroup_taskset *tset);
599 void (*fork)(struct task_struct *task);
600 void (*exit)(struct cgroup_subsys_state *css,
601 struct cgroup_subsys_state *old_css,
602 struct task_struct *task);
603 void (*bind)(struct cgroup_subsys_state *root_css);
609 * True if this subsys uses ID. ID is not available before cgroup_init()
610 * (not available in early_init time.)
615 * If %false, this subsystem is properly hierarchical -
616 * configuration, resource accounting and restriction on a parent
617 * cgroup cover those of its children. If %true, hierarchy support
618 * is broken in some ways - some subsystems ignore hierarchy
619 * completely while others are only implemented half-way.
621 * It's now disallowed to create nested cgroups if the subsystem is
622 * broken and cgroup core will emit a warning message on such
623 * cases. Eventually, all subsystems will be made properly
624 * hierarchical and this will go away.
626 bool broken_hierarchy;
627 bool warned_broken_hierarchy;
629 #define MAX_CGROUP_TYPE_NAMELEN 32
633 * Link to parent, and list entry in parent's children.
634 * Protected by cgroup_lock()
636 struct cgroupfs_root *root;
637 struct list_head sibling;
638 /* used when use_id == true */
642 /* list of cftype_sets */
643 struct list_head cftsets;
645 /* base cftypes, automatically [de]registered with subsys itself */
646 struct cftype *base_cftypes;
647 struct cftype_set base_cftset;
649 /* should be defined only by modular subsystems */
650 struct module *module;
653 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
654 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
655 #include <linux/cgroup_subsys.h>
656 #undef IS_SUBSYS_ENABLED
660 * css_parent - find the parent css
661 * @css: the target cgroup_subsys_state
663 * Return the parent css of @css. This function is guaranteed to return
664 * non-NULL parent as long as @css isn't the root.
667 struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css)
669 struct cgroup *parent_cgrp = css->cgroup->parent;
675 return parent_cgrp->subsys[css->ss->subsys_id];
677 return &parent_cgrp->dummy_css;
681 * task_css_set_check - obtain a task's css_set with extra access conditions
682 * @task: the task to obtain css_set for
683 * @__c: extra condition expression to be passed to rcu_dereference_check()
685 * A task's css_set is RCU protected, initialized and exited while holding
686 * task_lock(), and can only be modified while holding both cgroup_mutex
687 * and task_lock() while the task is alive. This macro verifies that the
688 * caller is inside proper critical section and returns @task's css_set.
690 * The caller can also specify additional allowed conditions via @__c, such
691 * as locks used during the cgroup_subsys::attach() methods.
693 #ifdef CONFIG_PROVE_RCU
694 extern struct mutex cgroup_mutex;
695 #define task_css_set_check(task, __c) \
696 rcu_dereference_check((task)->cgroups, \
697 lockdep_is_held(&(task)->alloc_lock) || \
698 lockdep_is_held(&cgroup_mutex) || (__c))
700 #define task_css_set_check(task, __c) \
701 rcu_dereference((task)->cgroups)
705 * task_css_check - obtain css for (task, subsys) w/ extra access conds
706 * @task: the target task
707 * @subsys_id: the target subsystem ID
708 * @__c: extra condition expression to be passed to rcu_dereference_check()
710 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
711 * synchronization rules are the same as task_css_set_check().
713 #define task_css_check(task, subsys_id, __c) \
714 task_css_set_check((task), (__c))->subsys[(subsys_id)]
717 * task_css_set - obtain a task's css_set
718 * @task: the task to obtain css_set for
720 * See task_css_set_check().
722 static inline struct css_set *task_css_set(struct task_struct *task)
724 return task_css_set_check(task, false);
728 * task_css - obtain css for (task, subsys)
729 * @task: the target task
730 * @subsys_id: the target subsystem ID
732 * See task_css_check().
734 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
737 return task_css_check(task, subsys_id, false);
740 static inline struct cgroup *task_cgroup(struct task_struct *task,
743 return task_css(task, subsys_id)->cgroup;
747 * cgroup_from_id - lookup cgroup by id
748 * @ss: cgroup subsys to be looked into
751 * Returns the cgroup if there's valid one with @id, otherwise returns NULL.
752 * Should be called under rcu_read_lock().
754 static inline struct cgroup *cgroup_from_id(struct cgroup_subsys *ss, int id)
756 #ifdef CONFIG_PROVE_RCU
757 rcu_lockdep_assert(rcu_read_lock_held() ||
758 lockdep_is_held(&cgroup_mutex),
759 "cgroup_from_id() needs proper protection");
761 return idr_find(&ss->root->cgroup_idr, id);
764 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
765 struct cgroup_subsys_state *parent);
768 * css_for_each_child - iterate through children of a css
769 * @pos: the css * to use as the loop cursor
770 * @parent: css whose children to walk
772 * Walk @parent's children. Must be called under rcu_read_lock(). A child
773 * css which hasn't finished ->css_online() or already has finished
774 * ->css_offline() may show up during traversal and it's each subsystem's
775 * responsibility to verify that each @pos is alive.
777 * If a subsystem synchronizes against the parent in its ->css_online() and
778 * before starting iterating, a css which finished ->css_online() is
779 * guaranteed to be visible in the future iterations.
781 * It is allowed to temporarily drop RCU read lock during iteration. The
782 * caller is responsible for ensuring that @pos remains accessible until
783 * the start of the next iteration by, for example, bumping the css refcnt.
785 #define css_for_each_child(pos, parent) \
786 for ((pos) = css_next_child(NULL, (parent)); (pos); \
787 (pos) = css_next_child((pos), (parent)))
789 struct cgroup_subsys_state *
790 css_next_descendant_pre(struct cgroup_subsys_state *pos,
791 struct cgroup_subsys_state *css);
793 struct cgroup_subsys_state *
794 css_rightmost_descendant(struct cgroup_subsys_state *pos);
797 * css_for_each_descendant_pre - pre-order walk of a css's descendants
798 * @pos: the css * to use as the loop cursor
799 * @root: css whose descendants to walk
801 * Walk @root's descendants. @root is included in the iteration and the
802 * first node to be visited. Must be called under rcu_read_lock(). A
803 * descendant css which hasn't finished ->css_online() or already has
804 * finished ->css_offline() may show up during traversal and it's each
805 * subsystem's responsibility to verify that each @pos is alive.
807 * If a subsystem synchronizes against the parent in its ->css_online() and
808 * before starting iterating, and synchronizes against @pos on each
809 * iteration, any descendant css which finished ->css_online() is
810 * guaranteed to be visible in the future iterations.
812 * In other words, the following guarantees that a descendant can't escape
813 * state updates of its ancestors.
817 * Lock @css's parent and @css;
818 * Inherit state from the parent;
822 * my_update_state(@css)
824 * css_for_each_descendant_pre(@pos, @css) {
827 * Update @css's state;
829 * Verify @pos is alive and inherit state from its parent;
834 * As long as the inheriting step, including checking the parent state, is
835 * enclosed inside @pos locking, double-locking the parent isn't necessary
836 * while inheriting. The state update to the parent is guaranteed to be
837 * visible by walking order and, as long as inheriting operations to the
838 * same @pos are atomic to each other, multiple updates racing each other
839 * still result in the correct state. It's guaranateed that at least one
840 * inheritance happens for any css after the latest update to its parent.
842 * If checking parent's state requires locking the parent, each inheriting
843 * iteration should lock and unlock both @pos->parent and @pos.
845 * Alternatively, a subsystem may choose to use a single global lock to
846 * synchronize ->css_online() and ->css_offline() against tree-walking
849 * It is allowed to temporarily drop RCU read lock during iteration. The
850 * caller is responsible for ensuring that @pos remains accessible until
851 * the start of the next iteration by, for example, bumping the css refcnt.
853 #define css_for_each_descendant_pre(pos, css) \
854 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
855 (pos) = css_next_descendant_pre((pos), (css)))
857 struct cgroup_subsys_state *
858 css_next_descendant_post(struct cgroup_subsys_state *pos,
859 struct cgroup_subsys_state *css);
862 * css_for_each_descendant_post - post-order walk of a css's descendants
863 * @pos: the css * to use as the loop cursor
864 * @css: css whose descendants to walk
866 * Similar to css_for_each_descendant_pre() but performs post-order
867 * traversal instead. @root is included in the iteration and the last
868 * node to be visited. Note that the walk visibility guarantee described
869 * in pre-order walk doesn't apply the same to post-order walks.
871 #define css_for_each_descendant_post(pos, css) \
872 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
873 (pos) = css_next_descendant_post((pos), (css)))
875 /* A css_task_iter should be treated as an opaque object */
876 struct css_task_iter {
877 struct cgroup_subsys_state *origin_css;
878 struct list_head *cset_link;
879 struct list_head *task;
882 void css_task_iter_start(struct cgroup_subsys_state *css,
883 struct css_task_iter *it);
884 struct task_struct *css_task_iter_next(struct css_task_iter *it);
885 void css_task_iter_end(struct css_task_iter *it);
887 int css_scan_tasks(struct cgroup_subsys_state *css,
888 bool (*test)(struct task_struct *, void *),
889 void (*process)(struct task_struct *, void *),
890 void *data, struct ptr_heap *heap);
892 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
893 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
896 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
897 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
898 * CSS ID is assigned at cgroup allocation (create) automatically
899 * and removed when subsys calls free_css_id() function. This is because
900 * the lifetime of cgroup_subsys_state is subsys's matter.
902 * Looking up and scanning function should be called under rcu_read_lock().
903 * Taking cgroup_mutex is not necessary for following calls.
904 * But the css returned by this routine can be "not populated yet" or "being
905 * destroyed". The caller should check css and cgroup's status.
909 * Typically Called at ->destroy(), or somewhere the subsys frees
910 * cgroup_subsys_state.
912 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
914 /* Find a cgroup_subsys_state which has given ID */
916 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
918 /* Returns true if root is ancestor of cg */
919 bool css_is_ancestor(struct cgroup_subsys_state *cg,
920 const struct cgroup_subsys_state *root);
922 /* Get id and depth of css */
923 unsigned short css_id(struct cgroup_subsys_state *css);
924 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
926 #else /* !CONFIG_CGROUPS */
928 static inline int cgroup_init_early(void) { return 0; }
929 static inline int cgroup_init(void) { return 0; }
930 static inline void cgroup_fork(struct task_struct *p) {}
931 static inline void cgroup_post_fork(struct task_struct *p) {}
932 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
934 static inline int cgroupstats_build(struct cgroupstats *stats,
935 struct dentry *dentry)
940 /* No cgroups - nothing to do */
941 static inline int cgroup_attach_task_all(struct task_struct *from,
942 struct task_struct *t)
947 #endif /* !CONFIG_CGROUPS */
949 #endif /* _LINUX_CGROUP_H */