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 /* For css percpu_ref killing and RCU-protected deletion */
229 struct rcu_head rcu_head
;
230 struct work_struct destroy_work
;
231 atomic_t css_kill_cnt
;
233 /* List of events which userspace want to receive */
234 struct list_head event_list
;
235 spinlock_t event_list_lock
;
237 /* directory xattrs */
238 struct simple_xattrs xattrs
;
241 #define MAX_CGROUP_ROOT_NAMELEN 64
243 /* cgroupfs_root->flags */
246 * Unfortunately, cgroup core and various controllers are riddled
247 * with idiosyncrasies and pointless options. The following flag,
248 * when set, will force sane behavior - some options are forced on,
249 * others are disallowed, and some controllers will change their
250 * hierarchical or other behaviors.
252 * The set of behaviors affected by this flag are still being
253 * determined and developed and the mount option for this flag is
254 * prefixed with __DEVEL__. The prefix will be dropped once we
255 * reach the point where all behaviors are compatible with the
256 * planned unified hierarchy, which will automatically turn on this
259 * The followings are the behaviors currently affected this flag.
261 * - Mount options "noprefix" and "clone_children" are disallowed.
262 * Also, cgroupfs file cgroup.clone_children is not created.
264 * - When mounting an existing superblock, mount options should
267 * - Remount is disallowed.
269 * - rename(2) is disallowed.
271 * - "tasks" is removed. Everything should be at process
272 * granularity. Use "cgroup.procs" instead.
274 * - "release_agent" and "notify_on_release" are removed.
275 * Replacement notification mechanism will be implemented.
277 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
278 * and take masks of ancestors with non-empty cpus/mems, instead of
279 * being moved to an ancestor.
281 * - cpuset: a task can be moved into an empty cpuset, and again it
282 * takes masks of ancestors.
284 * - memcg: use_hierarchy is on by default and the cgroup file for
285 * the flag is not created.
287 * - blkcg: blk-throttle becomes properly hierarchical.
289 CGRP_ROOT_SANE_BEHAVIOR
= (1 << 0),
291 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
292 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
294 /* mount options live below bit 16 */
295 CGRP_ROOT_OPTION_MASK
= (1 << 16) - 1,
297 CGRP_ROOT_SUBSYS_BOUND
= (1 << 16), /* subsystems finished binding */
301 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
302 * associated with a superblock to form an active hierarchy. This is
303 * internal to cgroup core. Don't access directly from controllers.
305 struct cgroupfs_root
{
306 struct super_block
*sb
;
308 /* The bitmask of subsystems attached to this hierarchy */
309 unsigned long subsys_mask
;
311 /* Unique id for this hierarchy. */
314 /* A list running through the attached subsystems */
315 struct list_head subsys_list
;
317 /* The root cgroup for this hierarchy */
318 struct cgroup top_cgroup
;
320 /* Tracks how many cgroups are currently defined in hierarchy.*/
321 int number_of_cgroups
;
323 /* A list running through the active hierarchies */
324 struct list_head root_list
;
326 /* Hierarchy-specific flags */
329 /* IDs for cgroups in this hierarchy */
330 struct idr cgroup_idr
;
332 /* The path to use for release notifications. */
333 char release_agent_path
[PATH_MAX
];
335 /* The name for this hierarchy - may be empty */
336 char name
[MAX_CGROUP_ROOT_NAMELEN
];
340 * A css_set is a structure holding pointers to a set of
341 * cgroup_subsys_state objects. This saves space in the task struct
342 * object and speeds up fork()/exit(), since a single inc/dec and a
343 * list_add()/del() can bump the reference count on the entire cgroup
349 /* Reference count */
353 * List running through all cgroup groups in the same hash
354 * slot. Protected by css_set_lock
356 struct hlist_node hlist
;
359 * List running through all tasks using this cgroup
360 * group. Protected by css_set_lock
362 struct list_head tasks
;
365 * List of cgrp_cset_links pointing at cgroups referenced from this
366 * css_set. Protected by css_set_lock.
368 struct list_head cgrp_links
;
371 * Set of subsystem states, one for each subsystem. This array
372 * is immutable after creation apart from the init_css_set
373 * during subsystem registration (at boot time) and modular subsystem
376 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
378 /* For RCU-protected deletion */
379 struct rcu_head rcu_head
;
383 * cgroup_map_cb is an abstract callback API for reporting map-valued
387 struct cgroup_map_cb
{
388 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
393 * struct cftype: handler definitions for cgroup control files
395 * When reading/writing to a file:
396 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
397 * - the 'cftype' of the file is file->f_dentry->d_fsdata
402 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
403 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
404 CFTYPE_INSANE
= (1 << 2), /* don't create if sane_behavior */
407 #define MAX_CFTYPE_NAME 64
411 * By convention, the name should begin with the name of the
412 * subsystem, followed by a period. Zero length string indicates
413 * end of cftype array.
415 char name
[MAX_CFTYPE_NAME
];
418 * If not 0, file mode is set to this value, otherwise it will
419 * be figured out automatically
424 * If non-zero, defines the maximum length of string that can
425 * be passed to write_string; defaults to 64
427 size_t max_write_len
;
433 * The subsys this file belongs to. Initialized automatically
434 * during registration. NULL for cgroup core files.
436 struct cgroup_subsys
*ss
;
438 int (*open
)(struct inode
*inode
, struct file
*file
);
439 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
441 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
443 * read_u64() is a shortcut for the common case of returning a
444 * single integer. Use it in place of read()
446 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
448 * read_s64() is a signed version of read_u64()
450 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
452 * read_map() is used for defining a map of key/value
453 * pairs. It should call cb->fill(cb, key, value) for each
454 * entry. The key/value pairs (and their ordering) should not
455 * change between reboots.
457 int (*read_map
)(struct cgroup
*cgrp
, struct cftype
*cft
,
458 struct cgroup_map_cb
*cb
);
460 * read_seq_string() is used for outputting a simple sequence
463 int (*read_seq_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
466 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
468 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
471 * write_u64() is a shortcut for the common case of accepting
472 * a single integer (as parsed by simple_strtoull) from
473 * userspace. Use in place of write(); return 0 or error.
475 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
477 * write_s64() is a signed version of write_u64()
479 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
482 * write_string() is passed a nul-terminated kernelspace
483 * buffer of maximum length determined by max_write_len.
484 * Returns 0 or -ve error code.
486 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
489 * trigger() callback can be used to get some kick from the
490 * userspace, when the actual string written is not important
491 * at all. The private field can be used to determine the
492 * kick type for multiplexing.
494 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
496 int (*release
)(struct inode
*inode
, struct file
*file
);
499 * register_event() callback will be used to add new userspace
500 * waiter for changes related to the cftype. Implement it if
501 * you want to provide this functionality. Use eventfd_signal()
502 * on eventfd to send notification to userspace.
504 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
505 struct eventfd_ctx
*eventfd
, const char *args
);
507 * unregister_event() callback will be called when userspace
508 * closes the eventfd or on cgroup removing.
509 * This callback must be implemented, if you want provide
510 * notification functionality.
512 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
513 struct eventfd_ctx
*eventfd
);
517 * cftype_sets describe cftypes belonging to a subsystem and are chained at
518 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
519 * terminated by zero length name.
522 struct list_head node
; /* chained at subsys->cftsets */
526 struct cgroup_scanner
{
528 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
529 void (*process_task
)(struct task_struct
*p
,
530 struct cgroup_scanner
*scan
);
531 struct ptr_heap
*heap
;
536 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
537 * function can be called as long as @cgrp is accessible.
539 static inline bool cgroup_sane_behavior(const struct cgroup
*cgrp
)
541 return cgrp
->root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
;
544 /* Caller should hold rcu_read_lock() */
545 static inline const char *cgroup_name(const struct cgroup
*cgrp
)
547 return rcu_dereference(cgrp
->name
)->name
;
550 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
551 int cgroup_rm_cftypes(struct cftype
*cfts
);
553 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
);
555 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
556 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
);
558 int cgroup_task_count(const struct cgroup
*cgrp
);
561 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
564 struct cgroup_taskset
;
565 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
566 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
567 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
568 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
571 * cgroup_taskset_for_each - iterate cgroup_taskset
572 * @task: the loop cursor
573 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
574 * @tset: taskset to iterate
576 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
577 for ((task) = cgroup_taskset_first((tset)); (task); \
578 (task) = cgroup_taskset_next((tset))) \
579 if (!(skip_cgrp) || \
580 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
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
;
671 return parent_cgrp
? parent_cgrp
->subsys
[css
->ss
->subsys_id
] : NULL
;
675 * cgroup_css - obtain a cgroup's css for the specified subsystem
676 * @cgrp: the cgroup of interest
677 * @subsys_id: the subsystem of interest
679 * Return @cgrp's css (cgroup_subsys_state) associated with @subsys_id.
681 static inline struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
684 return cgrp
->subsys
[subsys_id
];
688 * task_css_set_check - obtain a task's css_set with extra access conditions
689 * @task: the task to obtain css_set for
690 * @__c: extra condition expression to be passed to rcu_dereference_check()
692 * A task's css_set is RCU protected, initialized and exited while holding
693 * task_lock(), and can only be modified while holding both cgroup_mutex
694 * and task_lock() while the task is alive. This macro verifies that the
695 * caller is inside proper critical section and returns @task's css_set.
697 * The caller can also specify additional allowed conditions via @__c, such
698 * as locks used during the cgroup_subsys::attach() methods.
700 #ifdef CONFIG_PROVE_RCU
701 extern struct mutex cgroup_mutex
;
702 #define task_css_set_check(task, __c) \
703 rcu_dereference_check((task)->cgroups, \
704 lockdep_is_held(&(task)->alloc_lock) || \
705 lockdep_is_held(&cgroup_mutex) || (__c))
707 #define task_css_set_check(task, __c) \
708 rcu_dereference((task)->cgroups)
712 * task_css_check - obtain css for (task, subsys) w/ extra access conds
713 * @task: the target task
714 * @subsys_id: the target subsystem ID
715 * @__c: extra condition expression to be passed to rcu_dereference_check()
717 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
718 * synchronization rules are the same as task_css_set_check().
720 #define task_css_check(task, subsys_id, __c) \
721 task_css_set_check((task), (__c))->subsys[(subsys_id)]
724 * task_css_set - obtain a task's css_set
725 * @task: the task to obtain css_set for
727 * See task_css_set_check().
729 static inline struct css_set
*task_css_set(struct task_struct
*task
)
731 return task_css_set_check(task
, false);
735 * task_css - obtain css for (task, subsys)
736 * @task: the target task
737 * @subsys_id: the target subsystem ID
739 * See task_css_check().
741 static inline struct cgroup_subsys_state
*task_css(struct task_struct
*task
,
744 return task_css_check(task
, subsys_id
, false);
747 static inline struct cgroup
*task_cgroup(struct task_struct
*task
,
750 return task_css(task
, subsys_id
)->cgroup
;
754 * cgroup_from_id - lookup cgroup by id
755 * @ss: cgroup subsys to be looked into
758 * Returns the cgroup if there's valid one with @id, otherwise returns NULL.
759 * Should be called under rcu_read_lock().
761 static inline struct cgroup
*cgroup_from_id(struct cgroup_subsys
*ss
, int id
)
763 #ifdef CONFIG_PROVE_RCU
764 rcu_lockdep_assert(rcu_read_lock_held() ||
765 lockdep_is_held(&cgroup_mutex
),
766 "cgroup_from_id() needs proper protection");
768 return idr_find(&ss
->root
->cgroup_idr
, id
);
771 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
);
774 * cgroup_for_each_child - iterate through children of a cgroup
775 * @pos: the cgroup * to use as the loop cursor
776 * @cgrp: cgroup whose children to walk
778 * Walk @cgrp's children. Must be called under rcu_read_lock(). A child
779 * cgroup which hasn't finished ->css_online() or already has finished
780 * ->css_offline() may show up during traversal and it's each subsystem's
781 * responsibility to verify that each @pos is alive.
783 * If a subsystem synchronizes against the parent in its ->css_online() and
784 * before starting iterating, a cgroup which finished ->css_online() is
785 * guaranteed to be visible in the future iterations.
787 * It is allowed to temporarily drop RCU read lock during iteration. The
788 * caller is responsible for ensuring that @pos remains accessible until
789 * the start of the next iteration by, for example, bumping the css refcnt.
791 #define cgroup_for_each_child(pos, cgrp) \
792 for ((pos) = list_first_or_null_rcu(&(cgrp)->children, \
793 struct cgroup, sibling); \
794 (pos); (pos) = cgroup_next_sibling((pos)))
796 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
797 struct cgroup
*cgroup
);
798 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
);
801 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
802 * @pos: the cgroup * to use as the loop cursor
803 * @cgroup: cgroup whose descendants to walk
805 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
806 * descendant cgroup which hasn't finished ->css_online() or already has
807 * finished ->css_offline() may show up during traversal and it's each
808 * subsystem's responsibility to verify that each @pos is alive.
810 * If a subsystem synchronizes against the parent in its ->css_online() and
811 * before starting iterating, and synchronizes against @pos on each
812 * iteration, any descendant cgroup which finished ->css_online() is
813 * guaranteed to be visible in the future iterations.
815 * In other words, the following guarantees that a descendant can't escape
816 * state updates of its ancestors.
820 * Lock @cgrp->parent and @cgrp;
821 * Inherit state from @cgrp->parent;
825 * my_update_state(@cgrp)
828 * Update @cgrp's state;
831 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
833 * Verify @pos is alive and inherit state from @pos->parent;
838 * As long as the inheriting step, including checking the parent state, is
839 * enclosed inside @pos locking, double-locking the parent isn't necessary
840 * while inheriting. The state update to the parent is guaranteed to be
841 * visible by walking order and, as long as inheriting operations to the
842 * same @pos are atomic to each other, multiple updates racing each other
843 * still result in the correct state. It's guaranateed that at least one
844 * inheritance happens for any cgroup after the latest update to its
847 * If checking parent's state requires locking the parent, each inheriting
848 * iteration should lock and unlock both @pos->parent and @pos.
850 * Alternatively, a subsystem may choose to use a single global lock to
851 * synchronize ->css_online() and ->css_offline() against tree-walking
854 * It is allowed to temporarily drop RCU read lock during iteration. The
855 * caller is responsible for ensuring that @pos remains accessible until
856 * the start of the next iteration by, for example, bumping the css refcnt.
858 #define cgroup_for_each_descendant_pre(pos, cgroup) \
859 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
860 pos = cgroup_next_descendant_pre((pos), (cgroup)))
862 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
863 struct cgroup
*cgroup
);
866 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
867 * @pos: the cgroup * to use as the loop cursor
868 * @cgroup: cgroup whose descendants to walk
870 * Similar to cgroup_for_each_descendant_pre() but performs post-order
871 * traversal instead. Note that the walk visibility guarantee described in
872 * pre-order walk doesn't apply the same to post-order walks.
874 #define cgroup_for_each_descendant_post(pos, cgroup) \
875 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
876 pos = cgroup_next_descendant_post((pos), (cgroup)))
878 /* A cgroup_iter should be treated as an opaque object */
880 struct list_head
*cset_link
;
881 struct list_head
*task
;
885 * To iterate across the tasks in a cgroup:
887 * 1) call cgroup_iter_start to initialize an iterator
889 * 2) call cgroup_iter_next() to retrieve member tasks until it
890 * returns NULL or until you want to end the iteration
892 * 3) call cgroup_iter_end() to destroy the iterator.
894 * Or, call cgroup_scan_tasks() to iterate through every task in a
895 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
896 * the test_task() callback, but not while calling the process_task()
899 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
900 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
901 struct cgroup_iter
*it
);
902 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
903 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
904 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
905 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
);
908 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
909 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
910 * CSS ID is assigned at cgroup allocation (create) automatically
911 * and removed when subsys calls free_css_id() function. This is because
912 * the lifetime of cgroup_subsys_state is subsys's matter.
914 * Looking up and scanning function should be called under rcu_read_lock().
915 * Taking cgroup_mutex is not necessary for following calls.
916 * But the css returned by this routine can be "not populated yet" or "being
917 * destroyed". The caller should check css and cgroup's status.
921 * Typically Called at ->destroy(), or somewhere the subsys frees
922 * cgroup_subsys_state.
924 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
926 /* Find a cgroup_subsys_state which has given ID */
928 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
930 /* Returns true if root is ancestor of cg */
931 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
932 const struct cgroup_subsys_state
*root
);
934 /* Get id and depth of css */
935 unsigned short css_id(struct cgroup_subsys_state
*css
);
936 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
938 #else /* !CONFIG_CGROUPS */
940 static inline int cgroup_init_early(void) { return 0; }
941 static inline int cgroup_init(void) { return 0; }
942 static inline void cgroup_fork(struct task_struct
*p
) {}
943 static inline void cgroup_post_fork(struct task_struct
*p
) {}
944 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
946 static inline int cgroupstats_build(struct cgroupstats
*stats
,
947 struct dentry
*dentry
)
952 /* No cgroups - nothing to do */
953 static inline int cgroup_attach_task_all(struct task_struct
*from
,
954 struct task_struct
*t
)
959 #endif /* !CONFIG_CGROUPS */
961 #endif /* _LINUX_CGROUP_H */