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>
32 extern int cgroup_init_early(void);
33 extern int cgroup_init(void);
34 extern void cgroup_fork(struct task_struct
*p
);
35 extern void cgroup_post_fork(struct task_struct
*p
);
36 extern void cgroup_exit(struct task_struct
*p
);
37 extern int cgroupstats_build(struct cgroupstats
*stats
,
38 struct dentry
*dentry
);
40 extern int proc_cgroup_show(struct seq_file
*, void *);
42 /* define the enumeration of all cgroup subsystems */
43 #define SUBSYS(_x) _x ## _cgrp_id,
44 enum cgroup_subsys_id
{
45 #include <linux/cgroup_subsys.h>
50 /* Per-subsystem/per-cgroup state maintained by the system. */
51 struct cgroup_subsys_state
{
52 /* the cgroup that this css is attached to */
53 struct cgroup
*cgroup
;
55 /* the cgroup subsystem that this css is attached to */
56 struct cgroup_subsys
*ss
;
58 /* reference count - access via css_[try]get() and css_put() */
59 struct percpu_ref refcnt
;
62 struct cgroup_subsys_state
*parent
;
66 /* percpu_ref killing and RCU release */
67 struct rcu_head rcu_head
;
68 struct work_struct destroy_work
;
71 /* bits in struct cgroup_subsys_state flags field */
73 CSS_ROOT
= (1 << 0), /* this CSS is the root of the subsystem */
74 CSS_ONLINE
= (1 << 1), /* between ->css_online() and ->css_offline() */
78 * css_get - obtain a reference on the specified css
81 * The caller must already have a reference.
83 static inline void css_get(struct cgroup_subsys_state
*css
)
85 /* We don't need to reference count the root state */
86 if (!(css
->flags
& CSS_ROOT
))
87 percpu_ref_get(&css
->refcnt
);
91 * css_tryget - try to obtain a reference on the specified css
94 * Obtain a reference on @css if it's alive. The caller naturally needs to
95 * ensure that @css is accessible but doesn't have to be holding a
96 * reference on it - IOW, RCU protected access is good enough for this
97 * function. Returns %true if a reference count was successfully obtained;
100 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
102 if (css
->flags
& CSS_ROOT
)
104 return percpu_ref_tryget(&css
->refcnt
);
108 * css_put - put a css reference
111 * Put a reference obtained via css_get() and css_tryget().
113 static inline void css_put(struct cgroup_subsys_state
*css
)
115 if (!(css
->flags
& CSS_ROOT
))
116 percpu_ref_put(&css
->refcnt
);
119 /* bits in struct cgroup flags field */
121 /* Control Group is dead */
124 * Control Group has previously had a child cgroup or a task,
125 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
128 /* Control Group requires release notifications to userspace */
129 CGRP_NOTIFY_ON_RELEASE
,
131 * Clone the parent's configuration when creating a new child
132 * cpuset cgroup. For historical reasons, this option can be
133 * specified at mount time and thus is implemented here.
135 CGRP_CPUSET_CLONE_CHILDREN
,
136 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
141 unsigned long flags
; /* "unsigned long" so bitops work */
144 * idr allocated in-hierarchy ID.
146 * The ID of the root cgroup is always 0, and a new cgroup
147 * will be assigned with a smallest available ID.
149 * Allocating/Removing ID must be protected by cgroup_mutex.
153 /* the number of attached css's */
159 * We link our 'sibling' struct into our parent's 'children'.
160 * Our children link their 'sibling' into our 'children'.
162 struct list_head sibling
; /* my parent's children */
163 struct list_head children
; /* my children */
165 struct cgroup
*parent
; /* my parent */
166 struct kernfs_node
*kn
; /* cgroup kernfs entry */
169 * Monotonically increasing unique serial number which defines a
170 * uniform order among all cgroups. It's guaranteed that all
171 * ->children lists are in the ascending order of ->serial_nr.
172 * It's used to allow interrupting and resuming iterations.
176 /* the bitmask of subsystems enabled on the child cgroups */
177 unsigned long child_subsys_mask
;
179 /* Private pointers for each registered subsystem */
180 struct cgroup_subsys_state __rcu
*subsys
[CGROUP_SUBSYS_COUNT
];
182 struct cgroup_root
*root
;
185 * List of cgrp_cset_links pointing at css_sets with tasks in this
186 * cgroup. Protected by css_set_lock.
188 struct list_head cset_links
;
191 * On the default hierarchy, a css_set for a cgroup with some
192 * susbsys disabled will point to css's which are associated with
193 * the closest ancestor which has the subsys enabled. The
194 * following lists all css_sets which point to this cgroup's css
195 * for the given subsystem.
197 struct list_head e_csets
[CGROUP_SUBSYS_COUNT
];
200 * Linked list running through all cgroups that can
201 * potentially be reaped by the release agent. Protected by
204 struct list_head release_list
;
207 * list of pidlists, up to two for each namespace (one for procs, one
208 * for tasks); created on demand.
210 struct list_head pidlists
;
211 struct mutex pidlist_mutex
;
213 /* dummy css with NULL ->ss, points back to this cgroup */
214 struct cgroup_subsys_state dummy_css
;
216 /* For css percpu_ref killing and RCU-protected deletion */
217 struct rcu_head rcu_head
;
218 struct work_struct destroy_work
;
221 #define MAX_CGROUP_ROOT_NAMELEN 64
223 /* cgroup_root->flags */
226 * Unfortunately, cgroup core and various controllers are riddled
227 * with idiosyncrasies and pointless options. The following flag,
228 * when set, will force sane behavior - some options are forced on,
229 * others are disallowed, and some controllers will change their
230 * hierarchical or other behaviors.
232 * The set of behaviors affected by this flag are still being
233 * determined and developed and the mount option for this flag is
234 * prefixed with __DEVEL__. The prefix will be dropped once we
235 * reach the point where all behaviors are compatible with the
236 * planned unified hierarchy, which will automatically turn on this
239 * The followings are the behaviors currently affected this flag.
241 * - Mount options "noprefix", "xattr", "clone_children",
242 * "release_agent" and "name" are disallowed.
244 * - When mounting an existing superblock, mount options should
247 * - Remount is disallowed.
249 * - rename(2) is disallowed.
251 * - "tasks" is removed. Everything should be at process
252 * granularity. Use "cgroup.procs" instead.
254 * - "cgroup.procs" is not sorted. pids will be unique unless they
255 * got recycled inbetween reads.
257 * - "release_agent" and "notify_on_release" are removed.
258 * Replacement notification mechanism will be implemented.
260 * - "cgroup.clone_children" is removed.
262 * - If mount is requested with sane_behavior but without any
263 * subsystem, the default unified hierarchy is mounted.
265 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
266 * and take masks of ancestors with non-empty cpus/mems, instead of
267 * being moved to an ancestor.
269 * - cpuset: a task can be moved into an empty cpuset, and again it
270 * takes masks of ancestors.
272 * - memcg: use_hierarchy is on by default and the cgroup file for
273 * the flag is not created.
275 * - blkcg: blk-throttle becomes properly hierarchical.
277 CGRP_ROOT_SANE_BEHAVIOR
= (1 << 0),
279 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
280 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
282 /* mount options live below bit 16 */
283 CGRP_ROOT_OPTION_MASK
= (1 << 16) - 1,
287 * A cgroup_root represents the root of a cgroup hierarchy, and may be
288 * associated with a kernfs_root to form an active hierarchy. This is
289 * internal to cgroup core. Don't access directly from controllers.
292 struct kernfs_root
*kf_root
;
294 /* The bitmask of subsystems attached to this hierarchy */
295 unsigned long subsys_mask
;
297 /* Unique id for this hierarchy. */
300 /* The root cgroup. Root is destroyed on its release. */
303 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
306 /* A list running through the active hierarchies */
307 struct list_head root_list
;
309 /* Hierarchy-specific flags */
312 /* IDs for cgroups in this hierarchy */
313 struct idr cgroup_idr
;
315 /* The path to use for release notifications. */
316 char release_agent_path
[PATH_MAX
];
318 /* The name for this hierarchy - may be empty */
319 char name
[MAX_CGROUP_ROOT_NAMELEN
];
323 * A css_set is a structure holding pointers to a set of
324 * cgroup_subsys_state objects. This saves space in the task struct
325 * object and speeds up fork()/exit(), since a single inc/dec and a
326 * list_add()/del() can bump the reference count on the entire cgroup
332 /* Reference count */
336 * List running through all cgroup groups in the same hash
337 * slot. Protected by css_set_lock
339 struct hlist_node hlist
;
342 * Lists running through all tasks using this cgroup group.
343 * mg_tasks lists tasks which belong to this cset but are in the
344 * process of being migrated out or in. Protected by
345 * css_set_rwsem, but, during migration, once tasks are moved to
346 * mg_tasks, it can be read safely while holding cgroup_mutex.
348 struct list_head tasks
;
349 struct list_head mg_tasks
;
352 * List of cgrp_cset_links pointing at cgroups referenced from this
353 * css_set. Protected by css_set_lock.
355 struct list_head cgrp_links
;
358 * Set of subsystem states, one for each subsystem. This array is
359 * immutable after creation apart from the init_css_set during
360 * subsystem registration (at boot time).
362 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
365 * List of csets participating in the on-going migration either as
366 * source or destination. Protected by cgroup_mutex.
368 struct list_head mg_preload_node
;
369 struct list_head mg_node
;
372 * If this cset is acting as the source of migration the following
373 * two fields are set. mg_src_cgrp is the source cgroup of the
374 * on-going migration and mg_dst_cset is the destination cset the
375 * target tasks on this cset should be migrated to. Protected by
378 struct cgroup
*mg_src_cgrp
;
379 struct css_set
*mg_dst_cset
;
382 * On the default hierarhcy, ->subsys[ssid] may point to a css
383 * attached to an ancestor instead of the cgroup this css_set is
384 * associated with. The following node is anchored at
385 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
386 * iterate through all css's attached to a given cgroup.
388 struct list_head e_cset_node
[CGROUP_SUBSYS_COUNT
];
390 /* For RCU-protected deletion */
391 struct rcu_head rcu_head
;
395 * struct cftype: handler definitions for cgroup control files
397 * When reading/writing to a file:
398 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
399 * - the 'cftype' of the file is file->f_dentry->d_fsdata
404 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
405 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
406 CFTYPE_INSANE
= (1 << 2), /* don't create if sane_behavior */
407 CFTYPE_NO_PREFIX
= (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
408 CFTYPE_ONLY_ON_DFL
= (1 << 4), /* only on default hierarchy */
411 #define MAX_CFTYPE_NAME 64
415 * By convention, the name should begin with the name of the
416 * subsystem, followed by a period. Zero length string indicates
417 * end of cftype array.
419 char name
[MAX_CFTYPE_NAME
];
422 * If not 0, file mode is set to this value, otherwise it will
423 * be figured out automatically
428 * The maximum length of string, excluding trailing nul, that can
429 * be passed to write_string. If < PAGE_SIZE-1, PAGE_SIZE-1 is
432 size_t max_write_len
;
438 * Fields used for internal bookkeeping. Initialized automatically
439 * during registration.
441 struct cgroup_subsys
*ss
; /* NULL for cgroup core files */
442 struct list_head node
; /* anchored at ss->cfts */
443 struct kernfs_ops
*kf_ops
;
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 /* generic seq_file read interface */
456 int (*seq_show
)(struct seq_file
*sf
, void *v
);
458 /* optional ops, implement all or none */
459 void *(*seq_start
)(struct seq_file
*sf
, loff_t
*ppos
);
460 void *(*seq_next
)(struct seq_file
*sf
, void *v
, loff_t
*ppos
);
461 void (*seq_stop
)(struct seq_file
*sf
, void *v
);
464 * write_u64() is a shortcut for the common case of accepting
465 * a single integer (as parsed by simple_strtoull) from
466 * userspace. Use in place of write(); return 0 or error.
468 int (*write_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
471 * write_s64() is a signed version of write_u64()
473 int (*write_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
477 * write_string() is passed a nul-terminated kernelspace
478 * buffer of maximum length determined by max_write_len.
479 * Returns 0 or -ve error code.
481 int (*write_string
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
484 * trigger() callback can be used to get some kick from the
485 * userspace, when the actual string written is not important
486 * at all. The private field can be used to determine the
487 * kick type for multiplexing.
489 int (*trigger
)(struct cgroup_subsys_state
*css
, unsigned int event
);
491 #ifdef CONFIG_DEBUG_LOCK_ALLOC
492 struct lock_class_key lockdep_key
;
496 extern struct cgroup_root cgrp_dfl_root
;
498 static inline bool cgroup_on_dfl(const struct cgroup
*cgrp
)
500 return cgrp
->root
== &cgrp_dfl_root
;
504 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
505 * function can be called as long as @cgrp is accessible.
507 static inline bool cgroup_sane_behavior(const struct cgroup
*cgrp
)
509 return cgrp
->root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
;
512 /* no synchronization, the result can only be used as a hint */
513 static inline bool cgroup_has_tasks(struct cgroup
*cgrp
)
515 return !list_empty(&cgrp
->cset_links
);
518 /* returns ino associated with a cgroup, 0 indicates unmounted root */
519 static inline ino_t
cgroup_ino(struct cgroup
*cgrp
)
522 return cgrp
->kn
->ino
;
527 static inline struct cftype
*seq_cft(struct seq_file
*seq
)
529 struct kernfs_open_file
*of
= seq
->private;
534 struct cgroup_subsys_state
*seq_css(struct seq_file
*seq
);
537 * Name / path handling functions. All are thin wrappers around the kernfs
538 * counterparts and can be called under any context.
541 static inline int cgroup_name(struct cgroup
*cgrp
, char *buf
, size_t buflen
)
543 return kernfs_name(cgrp
->kn
, buf
, buflen
);
546 static inline char * __must_check
cgroup_path(struct cgroup
*cgrp
, char *buf
,
549 return kernfs_path(cgrp
->kn
, buf
, buflen
);
552 static inline void pr_cont_cgroup_name(struct cgroup
*cgrp
)
554 pr_cont_kernfs_name(cgrp
->kn
);
557 static inline void pr_cont_cgroup_path(struct cgroup
*cgrp
)
559 pr_cont_kernfs_path(cgrp
->kn
);
562 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
);
564 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
565 int cgroup_rm_cftypes(struct cftype
*cfts
);
567 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
);
570 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
573 struct cgroup_taskset
;
574 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
575 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
578 * cgroup_taskset_for_each - iterate cgroup_taskset
579 * @task: the loop cursor
580 * @tset: taskset to iterate
582 #define cgroup_taskset_for_each(task, tset) \
583 for ((task) = cgroup_taskset_first((tset)); (task); \
584 (task) = cgroup_taskset_next((tset)))
587 * Control Group subsystem type.
588 * See Documentation/cgroups/cgroups.txt for details
591 struct cgroup_subsys
{
592 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup_subsys_state
*parent_css
);
593 int (*css_online
)(struct cgroup_subsys_state
*css
);
594 void (*css_offline
)(struct cgroup_subsys_state
*css
);
595 void (*css_free
)(struct cgroup_subsys_state
*css
);
597 int (*can_attach
)(struct cgroup_subsys_state
*css
,
598 struct cgroup_taskset
*tset
);
599 void (*cancel_attach
)(struct cgroup_subsys_state
*css
,
600 struct cgroup_taskset
*tset
);
601 void (*attach
)(struct cgroup_subsys_state
*css
,
602 struct cgroup_taskset
*tset
);
603 void (*fork
)(struct task_struct
*task
);
604 void (*exit
)(struct cgroup_subsys_state
*css
,
605 struct cgroup_subsys_state
*old_css
,
606 struct task_struct
*task
);
607 void (*bind
)(struct cgroup_subsys_state
*root_css
);
613 * If %false, this subsystem is properly hierarchical -
614 * configuration, resource accounting and restriction on a parent
615 * cgroup cover those of its children. If %true, hierarchy support
616 * is broken in some ways - some subsystems ignore hierarchy
617 * completely while others are only implemented half-way.
619 * It's now disallowed to create nested cgroups if the subsystem is
620 * broken and cgroup core will emit a warning message on such
621 * cases. Eventually, all subsystems will be made properly
622 * hierarchical and this will go away.
624 bool broken_hierarchy
;
625 bool warned_broken_hierarchy
;
627 /* the following two fields are initialized automtically during boot */
629 #define MAX_CGROUP_TYPE_NAMELEN 32
632 /* link to parent, protected by cgroup_lock() */
633 struct cgroup_root
*root
;
636 * List of cftypes. Each entry is the first entry of an array
637 * terminated by zero length name.
639 struct list_head cfts
;
641 /* base cftypes, automatically registered with subsys itself */
642 struct cftype
*base_cftypes
;
645 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
646 #include <linux/cgroup_subsys.h>
650 * css_parent - find the parent css
651 * @css: the target cgroup_subsys_state
653 * Return the parent css of @css. This function is guaranteed to return
654 * non-NULL parent as long as @css isn't the root.
657 struct cgroup_subsys_state
*css_parent(struct cgroup_subsys_state
*css
)
663 * task_css_set_check - obtain a task's css_set with extra access conditions
664 * @task: the task to obtain css_set for
665 * @__c: extra condition expression to be passed to rcu_dereference_check()
667 * A task's css_set is RCU protected, initialized and exited while holding
668 * task_lock(), and can only be modified while holding both cgroup_mutex
669 * and task_lock() while the task is alive. This macro verifies that the
670 * caller is inside proper critical section and returns @task's css_set.
672 * The caller can also specify additional allowed conditions via @__c, such
673 * as locks used during the cgroup_subsys::attach() methods.
675 #ifdef CONFIG_PROVE_RCU
676 extern struct mutex cgroup_mutex
;
677 extern struct rw_semaphore css_set_rwsem
;
678 #define task_css_set_check(task, __c) \
679 rcu_dereference_check((task)->cgroups, \
680 lockdep_is_held(&cgroup_mutex) || \
681 lockdep_is_held(&css_set_rwsem) || \
682 ((task)->flags & PF_EXITING) || (__c))
684 #define task_css_set_check(task, __c) \
685 rcu_dereference((task)->cgroups)
689 * task_css_check - obtain css for (task, subsys) w/ extra access conds
690 * @task: the target task
691 * @subsys_id: the target subsystem ID
692 * @__c: extra condition expression to be passed to rcu_dereference_check()
694 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
695 * synchronization rules are the same as task_css_set_check().
697 #define task_css_check(task, subsys_id, __c) \
698 task_css_set_check((task), (__c))->subsys[(subsys_id)]
701 * task_css_set - obtain a task's css_set
702 * @task: the task to obtain css_set for
704 * See task_css_set_check().
706 static inline struct css_set
*task_css_set(struct task_struct
*task
)
708 return task_css_set_check(task
, false);
712 * task_css - obtain css for (task, subsys)
713 * @task: the target task
714 * @subsys_id: the target subsystem ID
716 * See task_css_check().
718 static inline struct cgroup_subsys_state
*task_css(struct task_struct
*task
,
721 return task_css_check(task
, subsys_id
, false);
724 static inline struct cgroup
*task_cgroup(struct task_struct
*task
,
727 return task_css(task
, subsys_id
)->cgroup
;
730 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
731 struct cgroup_subsys_state
*parent
);
733 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
);
736 * css_for_each_child - iterate through children of a css
737 * @pos: the css * to use as the loop cursor
738 * @parent: css whose children to walk
740 * Walk @parent's children. Must be called under rcu_read_lock(). A child
741 * css which hasn't finished ->css_online() or already has finished
742 * ->css_offline() may show up during traversal and it's each subsystem's
743 * responsibility to verify that each @pos is alive.
745 * If a subsystem synchronizes against the parent in its ->css_online() and
746 * before starting iterating, a css which finished ->css_online() is
747 * guaranteed to be visible in the future iterations.
749 * It is allowed to temporarily drop RCU read lock during iteration. The
750 * caller is responsible for ensuring that @pos remains accessible until
751 * the start of the next iteration by, for example, bumping the css refcnt.
753 #define css_for_each_child(pos, parent) \
754 for ((pos) = css_next_child(NULL, (parent)); (pos); \
755 (pos) = css_next_child((pos), (parent)))
757 struct cgroup_subsys_state
*
758 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
759 struct cgroup_subsys_state
*css
);
761 struct cgroup_subsys_state
*
762 css_rightmost_descendant(struct cgroup_subsys_state
*pos
);
765 * css_for_each_descendant_pre - pre-order walk of a css's descendants
766 * @pos: the css * to use as the loop cursor
767 * @root: css whose descendants to walk
769 * Walk @root's descendants. @root is included in the iteration and the
770 * first node to be visited. Must be called under rcu_read_lock(). A
771 * descendant css which hasn't finished ->css_online() or already has
772 * finished ->css_offline() may show up during traversal and it's each
773 * subsystem's responsibility to verify that each @pos is alive.
775 * If a subsystem synchronizes against the parent in its ->css_online() and
776 * before starting iterating, and synchronizes against @pos on each
777 * iteration, any descendant css which finished ->css_online() is
778 * guaranteed to be visible in the future iterations.
780 * In other words, the following guarantees that a descendant can't escape
781 * state updates of its ancestors.
785 * Lock @css's parent and @css;
786 * Inherit state from the parent;
790 * my_update_state(@css)
792 * css_for_each_descendant_pre(@pos, @css) {
795 * Update @css's state;
797 * Verify @pos is alive and inherit state from its parent;
802 * As long as the inheriting step, including checking the parent state, is
803 * enclosed inside @pos locking, double-locking the parent isn't necessary
804 * while inheriting. The state update to the parent is guaranteed to be
805 * visible by walking order and, as long as inheriting operations to the
806 * same @pos are atomic to each other, multiple updates racing each other
807 * still result in the correct state. It's guaranateed that at least one
808 * inheritance happens for any css after the latest update to its parent.
810 * If checking parent's state requires locking the parent, each inheriting
811 * iteration should lock and unlock both @pos->parent and @pos.
813 * Alternatively, a subsystem may choose to use a single global lock to
814 * synchronize ->css_online() and ->css_offline() against tree-walking
817 * It is allowed to temporarily drop RCU read lock during iteration. The
818 * caller is responsible for ensuring that @pos remains accessible until
819 * the start of the next iteration by, for example, bumping the css refcnt.
821 #define css_for_each_descendant_pre(pos, css) \
822 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
823 (pos) = css_next_descendant_pre((pos), (css)))
825 struct cgroup_subsys_state
*
826 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
827 struct cgroup_subsys_state
*css
);
830 * css_for_each_descendant_post - post-order walk of a css's descendants
831 * @pos: the css * to use as the loop cursor
832 * @css: css whose descendants to walk
834 * Similar to css_for_each_descendant_pre() but performs post-order
835 * traversal instead. @root is included in the iteration and the last
836 * node to be visited. Note that the walk visibility guarantee described
837 * in pre-order walk doesn't apply the same to post-order walks.
839 #define css_for_each_descendant_post(pos, css) \
840 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
841 (pos) = css_next_descendant_post((pos), (css)))
843 /* A css_task_iter should be treated as an opaque object */
844 struct css_task_iter
{
845 struct list_head
*cset_pos
;
846 struct list_head
*cset_head
;
848 struct list_head
*task_pos
;
849 struct list_head
*tasks_head
;
850 struct list_head
*mg_tasks_head
;
853 void css_task_iter_start(struct cgroup_subsys_state
*css
,
854 struct css_task_iter
*it
);
855 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
);
856 void css_task_iter_end(struct css_task_iter
*it
);
858 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
859 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
);
861 struct cgroup_subsys_state
*css_tryget_from_dir(struct dentry
*dentry
,
862 struct cgroup_subsys
*ss
);
864 #else /* !CONFIG_CGROUPS */
866 static inline int cgroup_init_early(void) { return 0; }
867 static inline int cgroup_init(void) { return 0; }
868 static inline void cgroup_fork(struct task_struct
*p
) {}
869 static inline void cgroup_post_fork(struct task_struct
*p
) {}
870 static inline void cgroup_exit(struct task_struct
*p
) {}
872 static inline int cgroupstats_build(struct cgroupstats
*stats
,
873 struct dentry
*dentry
)
878 /* No cgroups - nothing to do */
879 static inline int cgroup_attach_task_all(struct task_struct
*from
,
880 struct task_struct
*t
)
885 #endif /* !CONFIG_CGROUPS */
887 #endif /* _LINUX_CGROUP_H */