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_online - try to obtain a reference on the specified css if online
119 * Obtain a reference on @css if it's online. The caller naturally needs
120 * to ensure that @css is accessible but doesn't have to be holding a
121 * reference on it - IOW, RCU protected access is good enough for this
122 * function. Returns %true if a reference count was successfully obtained;
125 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
127 if (!(css->flags & CSS_NO_REF))
128 return percpu_ref_tryget_live(&css->refcnt);
133 * css_put - put a css reference
136 * Put a reference obtained via css_get() and css_tryget_online().
138 static inline void css_put(struct cgroup_subsys_state *css)
140 if (!(css->flags & CSS_NO_REF))
141 percpu_ref_put(&css->refcnt);
144 /* bits in struct cgroup flags field */
146 /* Control Group is dead */
149 * Control Group has previously had a child cgroup or a task,
150 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
153 /* Control Group requires release notifications to userspace */
154 CGRP_NOTIFY_ON_RELEASE,
156 * Clone the parent's configuration when creating a new child
157 * cpuset cgroup. For historical reasons, this option can be
158 * specified at mount time and thus is implemented here.
160 CGRP_CPUSET_CLONE_CHILDREN,
164 /* self css with NULL ->ss, points back to this cgroup */
165 struct cgroup_subsys_state self;
167 unsigned long flags; /* "unsigned long" so bitops work */
170 * idr allocated in-hierarchy ID.
172 * ID 0 is not used, the ID of the root cgroup is always 1, and a
173 * new cgroup will be assigned with a smallest available ID.
175 * Allocating/Removing ID must be protected by cgroup_mutex.
180 * If this cgroup contains any tasks, it contributes one to
181 * populated_cnt. All children with non-zero popuplated_cnt of
182 * their own contribute one. The count is zero iff there's no task
183 * in this cgroup or its subtree.
187 struct kernfs_node *kn; /* cgroup kernfs entry */
188 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
190 /* the bitmask of subsystems enabled on the child cgroups */
191 unsigned int child_subsys_mask;
193 /* Private pointers for each registered subsystem */
194 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
196 struct cgroup_root *root;
199 * List of cgrp_cset_links pointing at css_sets with tasks in this
200 * cgroup. Protected by css_set_lock.
202 struct list_head cset_links;
205 * On the default hierarchy, a css_set for a cgroup with some
206 * susbsys disabled will point to css's which are associated with
207 * the closest ancestor which has the subsys enabled. The
208 * following lists all css_sets which point to this cgroup's css
209 * for the given subsystem.
211 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
214 * Linked list running through all cgroups that can
215 * potentially be reaped by the release agent. Protected by
218 struct list_head release_list;
221 * list of pidlists, up to two for each namespace (one for procs, one
222 * for tasks); created on demand.
224 struct list_head pidlists;
225 struct mutex pidlist_mutex;
227 /* used to wait for offlining of csses */
228 wait_queue_head_t offline_waitq;
231 #define MAX_CGROUP_ROOT_NAMELEN 64
233 /* cgroup_root->flags */
236 * Unfortunately, cgroup core and various controllers are riddled
237 * with idiosyncrasies and pointless options. The following flag,
238 * when set, will force sane behavior - some options are forced on,
239 * others are disallowed, and some controllers will change their
240 * hierarchical or other behaviors.
242 * The set of behaviors affected by this flag are still being
243 * determined and developed and the mount option for this flag is
244 * prefixed with __DEVEL__. The prefix will be dropped once we
245 * reach the point where all behaviors are compatible with the
246 * planned unified hierarchy, which will automatically turn on this
249 * The followings are the behaviors currently affected this flag.
251 * - Mount options "noprefix", "xattr", "clone_children",
252 * "release_agent" and "name" are disallowed.
254 * - When mounting an existing superblock, mount options should
257 * - Remount is disallowed.
259 * - rename(2) is disallowed.
261 * - "tasks" is removed. Everything should be at process
262 * granularity. Use "cgroup.procs" instead.
264 * - "cgroup.procs" is not sorted. pids will be unique unless they
265 * got recycled inbetween reads.
267 * - "release_agent" and "notify_on_release" are removed.
268 * Replacement notification mechanism will be implemented.
270 * - "cgroup.clone_children" is removed.
272 * - "cgroup.subtree_populated" is available. Its value is 0 if
273 * the cgroup and its descendants contain no task; otherwise, 1.
274 * The file also generates kernfs notification which can be
275 * monitored through poll and [di]notify when the value of the
278 * - If mount is requested with sane_behavior but without any
279 * subsystem, the default unified hierarchy is mounted.
281 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
282 * and take masks of ancestors with non-empty cpus/mems, instead of
283 * being moved to an ancestor.
285 * - cpuset: a task can be moved into an empty cpuset, and again it
286 * takes masks of ancestors.
288 * - memcg: use_hierarchy is on by default and the cgroup file for
289 * the flag is not created.
291 * - blkcg: blk-throttle becomes properly hierarchical.
293 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
295 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
296 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
298 /* mount options live below bit 16 */
299 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
303 * A cgroup_root represents the root of a cgroup hierarchy, and may be
304 * associated with a kernfs_root to form an active hierarchy. This is
305 * internal to cgroup core. Don't access directly from controllers.
308 struct kernfs_root *kf_root;
310 /* The bitmask of subsystems attached to this hierarchy */
311 unsigned int subsys_mask;
313 /* Unique id for this hierarchy. */
316 /* The root cgroup. Root is destroyed on its release. */
319 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
322 /* A list running through the active hierarchies */
323 struct list_head root_list;
325 /* Hierarchy-specific flags */
328 /* IDs for cgroups in this hierarchy */
329 struct idr cgroup_idr;
331 /* The path to use for release notifications. */
332 char release_agent_path[PATH_MAX];
334 /* The name for this hierarchy - may be empty */
335 char name[MAX_CGROUP_ROOT_NAMELEN];
339 * A css_set is a structure holding pointers to a set of
340 * cgroup_subsys_state objects. This saves space in the task struct
341 * object and speeds up fork()/exit(), since a single inc/dec and a
342 * list_add()/del() can bump the reference count on the entire cgroup
348 /* Reference count */
352 * List running through all cgroup groups in the same hash
353 * slot. Protected by css_set_lock
355 struct hlist_node hlist;
358 * Lists running through all tasks using this cgroup group.
359 * mg_tasks lists tasks which belong to this cset but are in the
360 * process of being migrated out or in. Protected by
361 * css_set_rwsem, but, during migration, once tasks are moved to
362 * mg_tasks, it can be read safely while holding cgroup_mutex.
364 struct list_head tasks;
365 struct list_head mg_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;
373 /* the default cgroup associated with this css_set */
374 struct cgroup *dfl_cgrp;
377 * Set of subsystem states, one for each subsystem. This array is
378 * immutable after creation apart from the init_css_set during
379 * subsystem registration (at boot time).
381 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
384 * List of csets participating in the on-going migration either as
385 * source or destination. Protected by cgroup_mutex.
387 struct list_head mg_preload_node;
388 struct list_head mg_node;
391 * If this cset is acting as the source of migration the following
392 * two fields are set. mg_src_cgrp is the source cgroup of the
393 * on-going migration and mg_dst_cset is the destination cset the
394 * target tasks on this cset should be migrated to. Protected by
397 struct cgroup *mg_src_cgrp;
398 struct css_set *mg_dst_cset;
401 * On the default hierarhcy, ->subsys[ssid] may point to a css
402 * attached to an ancestor instead of the cgroup this css_set is
403 * associated with. The following node is anchored at
404 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
405 * iterate through all css's attached to a given cgroup.
407 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
409 /* For RCU-protected deletion */
410 struct rcu_head rcu_head;
414 * struct cftype: handler definitions for cgroup control files
416 * When reading/writing to a file:
417 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
418 * - the 'cftype' of the file is file->f_dentry->d_fsdata
423 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
424 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
425 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
426 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
427 CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */
430 #define MAX_CFTYPE_NAME 64
434 * By convention, the name should begin with the name of the
435 * subsystem, followed by a period. Zero length string indicates
436 * end of cftype array.
438 char name[MAX_CFTYPE_NAME];
441 * If not 0, file mode is set to this value, otherwise it will
442 * be figured out automatically
447 * The maximum length of string, excluding trailing nul, that can
448 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
450 size_t max_write_len;
456 * Fields used for internal bookkeeping. Initialized automatically
457 * during registration.
459 struct cgroup_subsys *ss; /* NULL for cgroup core files */
460 struct list_head node; /* anchored at ss->cfts */
461 struct kernfs_ops *kf_ops;
464 * read_u64() is a shortcut for the common case of returning a
465 * single integer. Use it in place of read()
467 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
469 * read_s64() is a signed version of read_u64()
471 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
473 /* generic seq_file read interface */
474 int (*seq_show)(struct seq_file *sf, void *v);
476 /* optional ops, implement all or none */
477 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
478 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
479 void (*seq_stop)(struct seq_file *sf, void *v);
482 * write_u64() is a shortcut for the common case of accepting
483 * a single integer (as parsed by simple_strtoull) from
484 * userspace. Use in place of write(); return 0 or error.
486 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
489 * write_s64() is a signed version of write_u64()
491 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
495 * write() is the generic write callback which maps directly to
496 * kernfs write operation and overrides all other operations.
497 * Maximum write size is determined by ->max_write_len. Use
498 * of_css/cft() to access the associated css and cft.
500 ssize_t (*write)(struct kernfs_open_file *of,
501 char *buf, size_t nbytes, loff_t off);
503 #ifdef CONFIG_DEBUG_LOCK_ALLOC
504 struct lock_class_key lockdep_key;
508 extern struct cgroup_root cgrp_dfl_root;
509 extern struct css_set init_css_set;
511 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
513 return cgrp->root == &cgrp_dfl_root;
517 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
518 * function can be called as long as @cgrp is accessible.
520 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
522 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
525 /* no synchronization, the result can only be used as a hint */
526 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
528 return !list_empty(&cgrp->cset_links);
531 /* returns ino associated with a cgroup, 0 indicates unmounted root */
532 static inline ino_t cgroup_ino(struct cgroup *cgrp)
535 return cgrp->kn->ino;
540 /* cft/css accessors for cftype->write() operation */
541 static inline struct cftype *of_cft(struct kernfs_open_file *of)
546 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
548 /* cft/css accessors for cftype->seq_*() operations */
549 static inline struct cftype *seq_cft(struct seq_file *seq)
551 return of_cft(seq->private);
554 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
556 return of_css(seq->private);
560 * Name / path handling functions. All are thin wrappers around the kernfs
561 * counterparts and can be called under any context.
564 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
566 return kernfs_name(cgrp->kn, buf, buflen);
569 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
572 return kernfs_path(cgrp->kn, buf, buflen);
575 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
577 pr_cont_kernfs_name(cgrp->kn);
580 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
582 pr_cont_kernfs_path(cgrp->kn);
585 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
587 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
588 int cgroup_rm_cftypes(struct cftype *cfts);
590 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
593 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
596 struct cgroup_taskset;
597 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
598 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
601 * cgroup_taskset_for_each - iterate cgroup_taskset
602 * @task: the loop cursor
603 * @tset: taskset to iterate
605 #define cgroup_taskset_for_each(task, tset) \
606 for ((task) = cgroup_taskset_first((tset)); (task); \
607 (task) = cgroup_taskset_next((tset)))
610 * Control Group subsystem type.
611 * See Documentation/cgroups/cgroups.txt for details
614 struct cgroup_subsys {
615 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
616 int (*css_online)(struct cgroup_subsys_state *css);
617 void (*css_offline)(struct cgroup_subsys_state *css);
618 void (*css_free)(struct cgroup_subsys_state *css);
620 int (*can_attach)(struct cgroup_subsys_state *css,
621 struct cgroup_taskset *tset);
622 void (*cancel_attach)(struct cgroup_subsys_state *css,
623 struct cgroup_taskset *tset);
624 void (*attach)(struct cgroup_subsys_state *css,
625 struct cgroup_taskset *tset);
626 void (*fork)(struct task_struct *task);
627 void (*exit)(struct cgroup_subsys_state *css,
628 struct cgroup_subsys_state *old_css,
629 struct task_struct *task);
630 void (*bind)(struct cgroup_subsys_state *root_css);
636 * If %false, this subsystem is properly hierarchical -
637 * configuration, resource accounting and restriction on a parent
638 * cgroup cover those of its children. If %true, hierarchy support
639 * is broken in some ways - some subsystems ignore hierarchy
640 * completely while others are only implemented half-way.
642 * It's now disallowed to create nested cgroups if the subsystem is
643 * broken and cgroup core will emit a warning message on such
644 * cases. Eventually, all subsystems will be made properly
645 * hierarchical and this will go away.
647 bool broken_hierarchy;
648 bool warned_broken_hierarchy;
650 /* the following two fields are initialized automtically during boot */
652 #define MAX_CGROUP_TYPE_NAMELEN 32
655 /* link to parent, protected by cgroup_lock() */
656 struct cgroup_root *root;
658 /* idr for css->id */
662 * List of cftypes. Each entry is the first entry of an array
663 * terminated by zero length name.
665 struct list_head cfts;
667 /* base cftypes, automatically registered with subsys itself */
668 struct cftype *base_cftypes;
671 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
672 #include <linux/cgroup_subsys.h>
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 extern struct rw_semaphore css_set_rwsem;
691 #define task_css_set_check(task, __c) \
692 rcu_dereference_check((task)->cgroups, \
693 lockdep_is_held(&cgroup_mutex) || \
694 lockdep_is_held(&css_set_rwsem) || \
695 ((task)->flags & PF_EXITING) || (__c))
697 #define task_css_set_check(task, __c) \
698 rcu_dereference((task)->cgroups)
702 * task_css_check - obtain css for (task, subsys) w/ extra access conds
703 * @task: the target task
704 * @subsys_id: the target subsystem ID
705 * @__c: extra condition expression to be passed to rcu_dereference_check()
707 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
708 * synchronization rules are the same as task_css_set_check().
710 #define task_css_check(task, subsys_id, __c) \
711 task_css_set_check((task), (__c))->subsys[(subsys_id)]
714 * task_css_set - obtain a task's css_set
715 * @task: the task to obtain css_set for
717 * See task_css_set_check().
719 static inline struct css_set *task_css_set(struct task_struct *task)
721 return task_css_set_check(task, false);
725 * task_css - obtain css for (task, subsys)
726 * @task: the target task
727 * @subsys_id: the target subsystem ID
729 * See task_css_check().
731 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
734 return task_css_check(task, subsys_id, false);
738 * task_css_is_root - test whether a task belongs to the root css
739 * @task: the target task
740 * @subsys_id: the target subsystem ID
742 * Test whether @task belongs to the root css on the specified subsystem.
743 * May be invoked in any context.
745 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
747 return task_css_check(task, subsys_id, true) ==
748 init_css_set.subsys[subsys_id];
751 static inline struct cgroup *task_cgroup(struct task_struct *task,
754 return task_css(task, subsys_id)->cgroup;
757 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
758 struct cgroup_subsys_state *parent);
760 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
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 *ss;
874 struct list_head *cset_pos;
875 struct list_head *cset_head;
877 struct list_head *task_pos;
878 struct list_head *tasks_head;
879 struct list_head *mg_tasks_head;
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 cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
888 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
890 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
891 struct cgroup_subsys *ss);
893 #else /* !CONFIG_CGROUPS */
895 static inline int cgroup_init_early(void) { return 0; }
896 static inline int cgroup_init(void) { return 0; }
897 static inline void cgroup_fork(struct task_struct *p) {}
898 static inline void cgroup_post_fork(struct task_struct *p) {}
899 static inline void cgroup_exit(struct task_struct *p) {}
901 static inline int cgroupstats_build(struct cgroupstats *stats,
902 struct dentry *dentry)
907 /* No cgroups - nothing to do */
908 static inline int cgroup_attach_task_all(struct task_struct *from,
909 struct task_struct *t)
914 #endif /* !CONFIG_CGROUPS */
916 #endif /* _LINUX_CGROUP_H */