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>
52 * Per-subsystem/per-cgroup state maintained by the system. This is the
53 * fundamental structural building block that controllers deal with.
55 * Fields marked with "PI:" are public and immutable and may be accessed
56 * directly without synchronization.
58 struct cgroup_subsys_state {
59 /* PI: the cgroup that this css is attached to */
60 struct cgroup *cgroup;
62 /* PI: the cgroup subsystem that this css is attached to */
63 struct cgroup_subsys *ss;
65 /* reference count - access via css_[try]get() and css_put() */
66 struct percpu_ref refcnt;
68 /* PI: the parent css */
69 struct cgroup_subsys_state *parent;
71 /* siblings list anchored at the parent's ->children */
72 struct list_head sibling;
73 struct list_head children;
76 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
77 * matching css can be looked up using css_from_id().
84 * Monotonically increasing unique serial number which defines a
85 * uniform order among all csses. It's guaranteed that all
86 * ->children lists are in the ascending order of ->serial_nr and
87 * used to allow interrupting and resuming iterations.
91 /* percpu_ref killing and RCU release */
92 struct rcu_head rcu_head;
93 struct work_struct destroy_work;
96 /* bits in struct cgroup_subsys_state flags field */
98 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
99 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
100 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
104 * css_get - obtain a reference on the specified css
107 * The caller must already have a reference.
109 static inline void css_get(struct cgroup_subsys_state *css)
111 if (!(css->flags & CSS_NO_REF))
112 percpu_ref_get(&css->refcnt);
116 * css_tryget - try to obtain a reference on the specified css
119 * Obtain a reference on @css unless it already has reached zero and is
120 * being released. This function doesn't care whether @css is on or
121 * offline. The caller naturally needs to ensure that @css is accessible
122 * but doesn't have to be holding a reference on it - IOW, RCU protected
123 * access is good enough for this function. Returns %true if a reference
124 * count was successfully obtained; %false otherwise.
126 static inline bool css_tryget(struct cgroup_subsys_state *css)
128 if (!(css->flags & CSS_NO_REF))
129 return percpu_ref_tryget(&css->refcnt);
134 * css_tryget_online - try to obtain a reference on the specified css if online
137 * Obtain a reference on @css if it's online. The caller naturally needs
138 * to ensure that @css is accessible but doesn't have to be holding a
139 * reference on it - IOW, RCU protected access is good enough for this
140 * function. Returns %true if a reference count was successfully obtained;
143 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
145 if (!(css->flags & CSS_NO_REF))
146 return percpu_ref_tryget_live(&css->refcnt);
151 * css_put - put a css reference
154 * Put a reference obtained via css_get() and css_tryget_online().
156 static inline void css_put(struct cgroup_subsys_state *css)
158 if (!(css->flags & CSS_NO_REF))
159 percpu_ref_put(&css->refcnt);
162 /* bits in struct cgroup flags field */
165 * Control Group has previously had a child cgroup or a task,
166 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
169 /* Control Group requires release notifications to userspace */
170 CGRP_NOTIFY_ON_RELEASE,
172 * Clone the parent's configuration when creating a new child
173 * cpuset cgroup. For historical reasons, this option can be
174 * specified at mount time and thus is implemented here.
176 CGRP_CPUSET_CLONE_CHILDREN,
180 /* self css with NULL ->ss, points back to this cgroup */
181 struct cgroup_subsys_state self;
183 unsigned long flags; /* "unsigned long" so bitops work */
186 * idr allocated in-hierarchy ID.
188 * ID 0 is not used, the ID of the root cgroup is always 1, and a
189 * new cgroup will be assigned with a smallest available ID.
191 * Allocating/Removing ID must be protected by cgroup_mutex.
196 * If this cgroup contains any tasks, it contributes one to
197 * populated_cnt. All children with non-zero popuplated_cnt of
198 * their own contribute one. The count is zero iff there's no task
199 * in this cgroup or its subtree.
203 struct kernfs_node *kn; /* cgroup kernfs entry */
204 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
207 * The bitmask of subsystems enabled on the child cgroups.
208 * ->subtree_control is the one configured through
209 * "cgroup.subtree_control" while ->child_subsys_mask is the
210 * effective one which may have more subsystems enabled.
211 * Controller knobs are made available iff it's enabled in
214 unsigned int subtree_control;
215 unsigned int child_subsys_mask;
217 /* Private pointers for each registered subsystem */
218 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
220 struct cgroup_root *root;
223 * List of cgrp_cset_links pointing at css_sets with tasks in this
224 * cgroup. Protected by css_set_lock.
226 struct list_head cset_links;
229 * On the default hierarchy, a css_set for a cgroup with some
230 * susbsys disabled will point to css's which are associated with
231 * the closest ancestor which has the subsys enabled. The
232 * following lists all css_sets which point to this cgroup's css
233 * for the given subsystem.
235 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
238 * Linked list running through all cgroups that can
239 * potentially be reaped by the release agent. Protected by
242 struct list_head release_list;
245 * list of pidlists, up to two for each namespace (one for procs, one
246 * for tasks); created on demand.
248 struct list_head pidlists;
249 struct mutex pidlist_mutex;
251 /* used to wait for offlining of csses */
252 wait_queue_head_t offline_waitq;
255 #define MAX_CGROUP_ROOT_NAMELEN 64
257 /* cgroup_root->flags */
260 * Unfortunately, cgroup core and various controllers are riddled
261 * with idiosyncrasies and pointless options. The following flag,
262 * when set, will force sane behavior - some options are forced on,
263 * others are disallowed, and some controllers will change their
264 * hierarchical or other behaviors.
266 * The set of behaviors affected by this flag are still being
267 * determined and developed and the mount option for this flag is
268 * prefixed with __DEVEL__. The prefix will be dropped once we
269 * reach the point where all behaviors are compatible with the
270 * planned unified hierarchy, which will automatically turn on this
273 * The followings are the behaviors currently affected this flag.
275 * - Mount options "noprefix", "xattr", "clone_children",
276 * "release_agent" and "name" are disallowed.
278 * - When mounting an existing superblock, mount options should
281 * - Remount is disallowed.
283 * - rename(2) is disallowed.
285 * - "tasks" is removed. Everything should be at process
286 * granularity. Use "cgroup.procs" instead.
288 * - "cgroup.procs" is not sorted. pids will be unique unless they
289 * got recycled inbetween reads.
291 * - "release_agent" and "notify_on_release" are removed.
292 * Replacement notification mechanism will be implemented.
294 * - "cgroup.clone_children" is removed.
296 * - "cgroup.subtree_populated" is available. Its value is 0 if
297 * the cgroup and its descendants contain no task; otherwise, 1.
298 * The file also generates kernfs notification which can be
299 * monitored through poll and [di]notify when the value of the
302 * - If mount is requested with sane_behavior but without any
303 * subsystem, the default unified hierarchy is mounted.
305 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
306 * and take masks of ancestors with non-empty cpus/mems, instead of
307 * being moved to an ancestor.
309 * - cpuset: a task can be moved into an empty cpuset, and again it
310 * takes masks of ancestors.
312 * - memcg: use_hierarchy is on by default and the cgroup file for
313 * the flag is not created.
315 * - blkcg: blk-throttle becomes properly hierarchical.
317 * - debug: disallowed on the default hierarchy.
319 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
321 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
322 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
324 /* mount options live below bit 16 */
325 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
329 * A cgroup_root represents the root of a cgroup hierarchy, and may be
330 * associated with a kernfs_root to form an active hierarchy. This is
331 * internal to cgroup core. Don't access directly from controllers.
334 struct kernfs_root *kf_root;
336 /* The bitmask of subsystems attached to this hierarchy */
337 unsigned int subsys_mask;
339 /* Unique id for this hierarchy. */
342 /* The root cgroup. Root is destroyed on its release. */
345 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
348 /* A list running through the active hierarchies */
349 struct list_head root_list;
351 /* Hierarchy-specific flags */
354 /* IDs for cgroups in this hierarchy */
355 struct idr cgroup_idr;
357 /* The path to use for release notifications. */
358 char release_agent_path[PATH_MAX];
360 /* The name for this hierarchy - may be empty */
361 char name[MAX_CGROUP_ROOT_NAMELEN];
365 * A css_set is a structure holding pointers to a set of
366 * cgroup_subsys_state objects. This saves space in the task struct
367 * object and speeds up fork()/exit(), since a single inc/dec and a
368 * list_add()/del() can bump the reference count on the entire cgroup
374 /* Reference count */
378 * List running through all cgroup groups in the same hash
379 * slot. Protected by css_set_lock
381 struct hlist_node hlist;
384 * Lists running through all tasks using this cgroup group.
385 * mg_tasks lists tasks which belong to this cset but are in the
386 * process of being migrated out or in. Protected by
387 * css_set_rwsem, but, during migration, once tasks are moved to
388 * mg_tasks, it can be read safely while holding cgroup_mutex.
390 struct list_head tasks;
391 struct list_head mg_tasks;
394 * List of cgrp_cset_links pointing at cgroups referenced from this
395 * css_set. Protected by css_set_lock.
397 struct list_head cgrp_links;
399 /* the default cgroup associated with this css_set */
400 struct cgroup *dfl_cgrp;
403 * Set of subsystem states, one for each subsystem. This array is
404 * immutable after creation apart from the init_css_set during
405 * subsystem registration (at boot time).
407 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
410 * List of csets participating in the on-going migration either as
411 * source or destination. Protected by cgroup_mutex.
413 struct list_head mg_preload_node;
414 struct list_head mg_node;
417 * If this cset is acting as the source of migration the following
418 * two fields are set. mg_src_cgrp is the source cgroup of the
419 * on-going migration and mg_dst_cset is the destination cset the
420 * target tasks on this cset should be migrated to. Protected by
423 struct cgroup *mg_src_cgrp;
424 struct css_set *mg_dst_cset;
427 * On the default hierarhcy, ->subsys[ssid] may point to a css
428 * attached to an ancestor instead of the cgroup this css_set is
429 * associated with. The following node is anchored at
430 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
431 * iterate through all css's attached to a given cgroup.
433 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
435 /* For RCU-protected deletion */
436 struct rcu_head rcu_head;
440 * struct cftype: handler definitions for cgroup control files
442 * When reading/writing to a file:
443 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
444 * - the 'cftype' of the file is file->f_dentry->d_fsdata
449 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
450 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
451 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
452 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
453 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
456 #define MAX_CFTYPE_NAME 64
460 * By convention, the name should begin with the name of the
461 * subsystem, followed by a period. Zero length string indicates
462 * end of cftype array.
464 char name[MAX_CFTYPE_NAME];
467 * If not 0, file mode is set to this value, otherwise it will
468 * be figured out automatically
473 * The maximum length of string, excluding trailing nul, that can
474 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
476 size_t max_write_len;
482 * Fields used for internal bookkeeping. Initialized automatically
483 * during registration.
485 struct cgroup_subsys *ss; /* NULL for cgroup core files */
486 struct list_head node; /* anchored at ss->cfts */
487 struct kernfs_ops *kf_ops;
490 * read_u64() is a shortcut for the common case of returning a
491 * single integer. Use it in place of read()
493 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
495 * read_s64() is a signed version of read_u64()
497 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
499 /* generic seq_file read interface */
500 int (*seq_show)(struct seq_file *sf, void *v);
502 /* optional ops, implement all or none */
503 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
504 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
505 void (*seq_stop)(struct seq_file *sf, void *v);
508 * write_u64() is a shortcut for the common case of accepting
509 * a single integer (as parsed by simple_strtoull) from
510 * userspace. Use in place of write(); return 0 or error.
512 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
515 * write_s64() is a signed version of write_u64()
517 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
521 * write() is the generic write callback which maps directly to
522 * kernfs write operation and overrides all other operations.
523 * Maximum write size is determined by ->max_write_len. Use
524 * of_css/cft() to access the associated css and cft.
526 ssize_t (*write)(struct kernfs_open_file *of,
527 char *buf, size_t nbytes, loff_t off);
529 #ifdef CONFIG_DEBUG_LOCK_ALLOC
530 struct lock_class_key lockdep_key;
534 extern struct cgroup_root cgrp_dfl_root;
535 extern struct css_set init_css_set;
537 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
539 return cgrp->root == &cgrp_dfl_root;
543 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
544 * function can be called as long as @cgrp is accessible.
546 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
548 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
551 /* no synchronization, the result can only be used as a hint */
552 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
554 return !list_empty(&cgrp->cset_links);
557 /* returns ino associated with a cgroup, 0 indicates unmounted root */
558 static inline ino_t cgroup_ino(struct cgroup *cgrp)
561 return cgrp->kn->ino;
566 /* cft/css accessors for cftype->write() operation */
567 static inline struct cftype *of_cft(struct kernfs_open_file *of)
572 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
574 /* cft/css accessors for cftype->seq_*() operations */
575 static inline struct cftype *seq_cft(struct seq_file *seq)
577 return of_cft(seq->private);
580 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
582 return of_css(seq->private);
586 * Name / path handling functions. All are thin wrappers around the kernfs
587 * counterparts and can be called under any context.
590 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
592 return kernfs_name(cgrp->kn, buf, buflen);
595 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
598 return kernfs_path(cgrp->kn, buf, buflen);
601 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
603 pr_cont_kernfs_name(cgrp->kn);
606 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
608 pr_cont_kernfs_path(cgrp->kn);
611 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
613 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
614 int cgroup_rm_cftypes(struct cftype *cfts);
616 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
619 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
622 struct cgroup_taskset;
623 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
624 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
627 * cgroup_taskset_for_each - iterate cgroup_taskset
628 * @task: the loop cursor
629 * @tset: taskset to iterate
631 #define cgroup_taskset_for_each(task, tset) \
632 for ((task) = cgroup_taskset_first((tset)); (task); \
633 (task) = cgroup_taskset_next((tset)))
636 * Control Group subsystem type.
637 * See Documentation/cgroups/cgroups.txt for details
640 struct cgroup_subsys {
641 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
642 int (*css_online)(struct cgroup_subsys_state *css);
643 void (*css_offline)(struct cgroup_subsys_state *css);
644 void (*css_free)(struct cgroup_subsys_state *css);
645 void (*css_reset)(struct cgroup_subsys_state *css);
647 int (*can_attach)(struct cgroup_subsys_state *css,
648 struct cgroup_taskset *tset);
649 void (*cancel_attach)(struct cgroup_subsys_state *css,
650 struct cgroup_taskset *tset);
651 void (*attach)(struct cgroup_subsys_state *css,
652 struct cgroup_taskset *tset);
653 void (*fork)(struct task_struct *task);
654 void (*exit)(struct cgroup_subsys_state *css,
655 struct cgroup_subsys_state *old_css,
656 struct task_struct *task);
657 void (*bind)(struct cgroup_subsys_state *root_css);
663 * If %false, this subsystem is properly hierarchical -
664 * configuration, resource accounting and restriction on a parent
665 * cgroup cover those of its children. If %true, hierarchy support
666 * is broken in some ways - some subsystems ignore hierarchy
667 * completely while others are only implemented half-way.
669 * It's now disallowed to create nested cgroups if the subsystem is
670 * broken and cgroup core will emit a warning message on such
671 * cases. Eventually, all subsystems will be made properly
672 * hierarchical and this will go away.
674 bool broken_hierarchy;
675 bool warned_broken_hierarchy;
677 /* the following two fields are initialized automtically during boot */
679 #define MAX_CGROUP_TYPE_NAMELEN 32
682 /* link to parent, protected by cgroup_lock() */
683 struct cgroup_root *root;
685 /* idr for css->id */
689 * List of cftypes. Each entry is the first entry of an array
690 * terminated by zero length name.
692 struct list_head cfts;
694 /* base cftypes, automatically registered with subsys itself */
695 struct cftype *base_cftypes;
698 * A subsystem may depend on other subsystems. When such subsystem
699 * is enabled on a cgroup, the depended-upon subsystems are enabled
700 * together if available. Subsystems enabled due to dependency are
701 * not visible to userland until explicitly enabled. The following
702 * specifies the mask of subsystems that this one depends on.
704 unsigned int depends_on;
707 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
708 #include <linux/cgroup_subsys.h>
712 * task_css_set_check - obtain a task's css_set with extra access conditions
713 * @task: the task to obtain css_set for
714 * @__c: extra condition expression to be passed to rcu_dereference_check()
716 * A task's css_set is RCU protected, initialized and exited while holding
717 * task_lock(), and can only be modified while holding both cgroup_mutex
718 * and task_lock() while the task is alive. This macro verifies that the
719 * caller is inside proper critical section and returns @task's css_set.
721 * The caller can also specify additional allowed conditions via @__c, such
722 * as locks used during the cgroup_subsys::attach() methods.
724 #ifdef CONFIG_PROVE_RCU
725 extern struct mutex cgroup_mutex;
726 extern struct rw_semaphore css_set_rwsem;
727 #define task_css_set_check(task, __c) \
728 rcu_dereference_check((task)->cgroups, \
729 lockdep_is_held(&cgroup_mutex) || \
730 lockdep_is_held(&css_set_rwsem) || \
731 ((task)->flags & PF_EXITING) || (__c))
733 #define task_css_set_check(task, __c) \
734 rcu_dereference((task)->cgroups)
738 * task_css_check - obtain css for (task, subsys) w/ extra access conds
739 * @task: the target task
740 * @subsys_id: the target subsystem ID
741 * @__c: extra condition expression to be passed to rcu_dereference_check()
743 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
744 * synchronization rules are the same as task_css_set_check().
746 #define task_css_check(task, subsys_id, __c) \
747 task_css_set_check((task), (__c))->subsys[(subsys_id)]
750 * task_css_set - obtain a task's css_set
751 * @task: the task to obtain css_set for
753 * See task_css_set_check().
755 static inline struct css_set *task_css_set(struct task_struct *task)
757 return task_css_set_check(task, false);
761 * task_css - obtain css for (task, subsys)
762 * @task: the target task
763 * @subsys_id: the target subsystem ID
765 * See task_css_check().
767 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
770 return task_css_check(task, subsys_id, false);
774 * task_css_is_root - test whether a task belongs to the root css
775 * @task: the target task
776 * @subsys_id: the target subsystem ID
778 * Test whether @task belongs to the root css on the specified subsystem.
779 * May be invoked in any context.
781 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
783 return task_css_check(task, subsys_id, true) ==
784 init_css_set.subsys[subsys_id];
787 static inline struct cgroup *task_cgroup(struct task_struct *task,
790 return task_css(task, subsys_id)->cgroup;
793 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
794 struct cgroup_subsys_state *parent);
796 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
799 * css_for_each_child - iterate through children of a css
800 * @pos: the css * to use as the loop cursor
801 * @parent: css whose children to walk
803 * Walk @parent's children. Must be called under rcu_read_lock().
805 * If a subsystem synchronizes ->css_online() and the start of iteration, a
806 * css which finished ->css_online() is guaranteed to be visible in the
807 * future iterations and will stay visible until the last reference is put.
808 * A css which hasn't finished ->css_online() or already finished
809 * ->css_offline() may show up during traversal. It's each subsystem's
810 * responsibility to synchronize against on/offlining.
812 * It is allowed to temporarily drop RCU read lock during iteration. The
813 * caller is responsible for ensuring that @pos remains accessible until
814 * the start of the next iteration by, for example, bumping the css refcnt.
816 #define css_for_each_child(pos, parent) \
817 for ((pos) = css_next_child(NULL, (parent)); (pos); \
818 (pos) = css_next_child((pos), (parent)))
820 struct cgroup_subsys_state *
821 css_next_descendant_pre(struct cgroup_subsys_state *pos,
822 struct cgroup_subsys_state *css);
824 struct cgroup_subsys_state *
825 css_rightmost_descendant(struct cgroup_subsys_state *pos);
828 * css_for_each_descendant_pre - pre-order walk of a css's descendants
829 * @pos: the css * to use as the loop cursor
830 * @root: css whose descendants to walk
832 * Walk @root's descendants. @root is included in the iteration and the
833 * first node to be visited. Must be called under rcu_read_lock().
835 * If a subsystem synchronizes ->css_online() and the start of iteration, a
836 * css which finished ->css_online() is guaranteed to be visible in the
837 * future iterations and will stay visible until the last reference is put.
838 * A css which hasn't finished ->css_online() or already finished
839 * ->css_offline() may show up during traversal. It's each subsystem's
840 * responsibility to synchronize against on/offlining.
842 * For example, the following guarantees that a descendant can't escape
843 * state updates of its ancestors.
847 * Lock @css's parent and @css;
848 * Inherit state from the parent;
852 * my_update_state(@css)
854 * css_for_each_descendant_pre(@pos, @css) {
857 * Update @css's state;
859 * Verify @pos is alive and inherit state from its parent;
864 * As long as the inheriting step, including checking the parent state, is
865 * enclosed inside @pos locking, double-locking the parent isn't necessary
866 * while inheriting. The state update to the parent is guaranteed to be
867 * visible by walking order and, as long as inheriting operations to the
868 * same @pos are atomic to each other, multiple updates racing each other
869 * still result in the correct state. It's guaranateed that at least one
870 * inheritance happens for any css after the latest update to its parent.
872 * If checking parent's state requires locking the parent, each inheriting
873 * iteration should lock and unlock both @pos->parent and @pos.
875 * Alternatively, a subsystem may choose to use a single global lock to
876 * synchronize ->css_online() and ->css_offline() against tree-walking
879 * It is allowed to temporarily drop RCU read lock during iteration. The
880 * caller is responsible for ensuring that @pos remains accessible until
881 * the start of the next iteration by, for example, bumping the css refcnt.
883 #define css_for_each_descendant_pre(pos, css) \
884 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
885 (pos) = css_next_descendant_pre((pos), (css)))
887 struct cgroup_subsys_state *
888 css_next_descendant_post(struct cgroup_subsys_state *pos,
889 struct cgroup_subsys_state *css);
892 * css_for_each_descendant_post - post-order walk of a css's descendants
893 * @pos: the css * to use as the loop cursor
894 * @css: css whose descendants to walk
896 * Similar to css_for_each_descendant_pre() but performs post-order
897 * traversal instead. @root is included in the iteration and the last
898 * node to be visited.
900 * If a subsystem synchronizes ->css_online() and the start of iteration, a
901 * css which finished ->css_online() is guaranteed to be visible in the
902 * future iterations and will stay visible until the last reference is put.
903 * A css which hasn't finished ->css_online() or already finished
904 * ->css_offline() may show up during traversal. It's each subsystem's
905 * responsibility to synchronize against on/offlining.
907 * Note that the walk visibility guarantee example described in pre-order
908 * walk doesn't apply the same to post-order walks.
910 #define css_for_each_descendant_post(pos, css) \
911 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
912 (pos) = css_next_descendant_post((pos), (css)))
914 bool css_has_online_children(struct cgroup_subsys_state *css);
916 /* A css_task_iter should be treated as an opaque object */
917 struct css_task_iter {
918 struct cgroup_subsys *ss;
920 struct list_head *cset_pos;
921 struct list_head *cset_head;
923 struct list_head *task_pos;
924 struct list_head *tasks_head;
925 struct list_head *mg_tasks_head;
928 void css_task_iter_start(struct cgroup_subsys_state *css,
929 struct css_task_iter *it);
930 struct task_struct *css_task_iter_next(struct css_task_iter *it);
931 void css_task_iter_end(struct css_task_iter *it);
933 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
934 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
936 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
937 struct cgroup_subsys *ss);
939 #else /* !CONFIG_CGROUPS */
941 static inline int cgroup_init_early(void) { return 0; }
942 static inline int cgroup_init(void) { return 0; }
943 static inline void cgroup_fork(struct task_struct *p) {}
944 static inline void cgroup_post_fork(struct task_struct *p) {}
945 static inline void cgroup_exit(struct task_struct *p) {}
947 static inline int cgroupstats_build(struct cgroupstats *stats,
948 struct dentry *dentry)
953 /* No cgroups - nothing to do */
954 static inline int cgroup_attach_task_all(struct task_struct *from,
955 struct task_struct *t)
960 #endif /* !CONFIG_CGROUPS */
962 #endif /* _LINUX_CGROUP_H */