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 struct cgroup_subsys_state
*parent
;
82 /* ID for this css, if possible */
83 struct css_id __rcu
*id
;
85 /* percpu_ref killing and RCU release */
86 struct rcu_head rcu_head
;
87 struct work_struct destroy_work
;
90 /* bits in struct cgroup_subsys_state flags field */
92 CSS_ROOT
= (1 << 0), /* this CSS is the root of the subsystem */
93 CSS_ONLINE
= (1 << 1), /* between ->css_online() and ->css_offline() */
97 * css_get - obtain a reference on the specified css
100 * The caller must already have a reference.
102 static inline void css_get(struct cgroup_subsys_state
*css
)
104 /* We don't need to reference count the root state */
105 if (!(css
->flags
& CSS_ROOT
))
106 percpu_ref_get(&css
->refcnt
);
110 * css_tryget - try to obtain a reference on the specified css
113 * Obtain a reference on @css if it's alive. The caller naturally needs to
114 * ensure that @css is accessible but doesn't have to be holding a
115 * reference on it - IOW, RCU protected access is good enough for this
116 * function. Returns %true if a reference count was successfully obtained;
119 static inline bool css_tryget(struct cgroup_subsys_state
*css
)
121 if (css
->flags
& CSS_ROOT
)
123 return percpu_ref_tryget(&css
->refcnt
);
127 * css_put - put a css reference
130 * Put a reference obtained via css_get() and css_tryget().
132 static inline void css_put(struct cgroup_subsys_state
*css
)
134 if (!(css
->flags
& CSS_ROOT
))
135 percpu_ref_put(&css
->refcnt
);
138 /* bits in struct cgroup flags field */
140 /* Control Group is dead */
143 * Control Group has previously had a child cgroup or a task,
144 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
147 /* Control Group requires release notifications to userspace */
148 CGRP_NOTIFY_ON_RELEASE
,
150 * Clone the parent's configuration when creating a new child
151 * cpuset cgroup. For historical reasons, this option can be
152 * specified at mount time and thus is implemented here.
154 CGRP_CPUSET_CLONE_CHILDREN
,
155 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
160 struct rcu_head rcu_head
;
165 unsigned long flags
; /* "unsigned long" so bitops work */
168 * idr allocated in-hierarchy ID.
170 * The ID of the root cgroup is always 0, and a new cgroup
171 * will be assigned with a smallest available ID.
175 /* the number of attached css's */
179 * We link our 'sibling' struct into our parent's 'children'.
180 * Our children link their 'sibling' into our 'children'.
182 struct list_head sibling
; /* my parent's children */
183 struct list_head children
; /* my children */
184 struct list_head files
; /* my files */
186 struct cgroup
*parent
; /* my parent */
187 struct dentry
*dentry
; /* cgroup fs entry, RCU protected */
190 * Monotonically increasing unique serial number which defines a
191 * uniform order among all cgroups. It's guaranteed that all
192 * ->children lists are in the ascending order of ->serial_nr.
193 * It's used to allow interrupting and resuming iterations.
198 * This is a copy of dentry->d_name, and it's needed because
199 * we can't use dentry->d_name in cgroup_path().
201 * You must acquire rcu_read_lock() to access cgrp->name, and
202 * the only place that can change it is rename(), which is
203 * protected by parent dir's i_mutex.
205 * Normally you should use cgroup_name() wrapper rather than
206 * access it directly.
208 struct cgroup_name __rcu
*name
;
210 /* Private pointers for each registered subsystem */
211 struct cgroup_subsys_state __rcu
*subsys
[CGROUP_SUBSYS_COUNT
];
213 struct cgroupfs_root
*root
;
216 * List of cgrp_cset_links pointing at css_sets with tasks in this
217 * cgroup. Protected by css_set_lock.
219 struct list_head cset_links
;
222 * Linked list running through all cgroups that can
223 * potentially be reaped by the release agent. Protected by
226 struct list_head release_list
;
229 * list of pidlists, up to two for each namespace (one for procs, one
230 * for tasks); created on demand.
232 struct list_head pidlists
;
233 struct mutex pidlist_mutex
;
235 /* dummy css with NULL ->ss, points back to this cgroup */
236 struct cgroup_subsys_state dummy_css
;
238 /* For css percpu_ref killing and RCU-protected deletion */
239 struct rcu_head rcu_head
;
240 struct work_struct destroy_work
;
242 /* List of events which userspace want to receive */
243 struct list_head event_list
;
244 spinlock_t event_list_lock
;
246 /* directory xattrs */
247 struct simple_xattrs xattrs
;
250 #define MAX_CGROUP_ROOT_NAMELEN 64
252 /* cgroupfs_root->flags */
255 * Unfortunately, cgroup core and various controllers are riddled
256 * with idiosyncrasies and pointless options. The following flag,
257 * when set, will force sane behavior - some options are forced on,
258 * others are disallowed, and some controllers will change their
259 * hierarchical or other behaviors.
261 * The set of behaviors affected by this flag are still being
262 * determined and developed and the mount option for this flag is
263 * prefixed with __DEVEL__. The prefix will be dropped once we
264 * reach the point where all behaviors are compatible with the
265 * planned unified hierarchy, which will automatically turn on this
268 * The followings are the behaviors currently affected this flag.
270 * - Mount options "noprefix" and "clone_children" are disallowed.
271 * Also, cgroupfs file cgroup.clone_children is not created.
273 * - When mounting an existing superblock, mount options should
276 * - Remount is disallowed.
278 * - rename(2) is disallowed.
280 * - "tasks" is removed. Everything should be at process
281 * granularity. Use "cgroup.procs" instead.
283 * - "release_agent" and "notify_on_release" are removed.
284 * Replacement notification mechanism will be implemented.
286 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
287 * and take masks of ancestors with non-empty cpus/mems, instead of
288 * being moved to an ancestor.
290 * - cpuset: a task can be moved into an empty cpuset, and again it
291 * takes masks of ancestors.
293 * - memcg: use_hierarchy is on by default and the cgroup file for
294 * the flag is not created.
296 * - blkcg: blk-throttle becomes properly hierarchical.
298 CGRP_ROOT_SANE_BEHAVIOR
= (1 << 0),
300 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
301 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
303 /* mount options live below bit 16 */
304 CGRP_ROOT_OPTION_MASK
= (1 << 16) - 1,
306 CGRP_ROOT_SUBSYS_BOUND
= (1 << 16), /* subsystems finished binding */
310 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
311 * associated with a superblock to form an active hierarchy. This is
312 * internal to cgroup core. Don't access directly from controllers.
314 struct cgroupfs_root
{
315 struct super_block
*sb
;
317 /* The bitmask of subsystems attached to this hierarchy */
318 unsigned long subsys_mask
;
320 /* Unique id for this hierarchy. */
323 /* A list running through the attached subsystems */
324 struct list_head subsys_list
;
326 /* The root cgroup for this hierarchy */
327 struct cgroup top_cgroup
;
329 /* Tracks how many cgroups are currently defined in hierarchy.*/
330 int number_of_cgroups
;
332 /* A list running through the active hierarchies */
333 struct list_head root_list
;
335 /* Hierarchy-specific flags */
338 /* IDs for cgroups in this hierarchy */
339 struct idr cgroup_idr
;
341 /* The path to use for release notifications. */
342 char release_agent_path
[PATH_MAX
];
344 /* The name for this hierarchy - may be empty */
345 char name
[MAX_CGROUP_ROOT_NAMELEN
];
349 * A css_set is a structure holding pointers to a set of
350 * cgroup_subsys_state objects. This saves space in the task struct
351 * object and speeds up fork()/exit(), since a single inc/dec and a
352 * list_add()/del() can bump the reference count on the entire cgroup
358 /* Reference count */
362 * List running through all cgroup groups in the same hash
363 * slot. Protected by css_set_lock
365 struct hlist_node hlist
;
368 * List running through all tasks using this cgroup
369 * group. Protected by css_set_lock
371 struct list_head tasks
;
374 * List of cgrp_cset_links pointing at cgroups referenced from this
375 * css_set. Protected by css_set_lock.
377 struct list_head cgrp_links
;
380 * Set of subsystem states, one for each subsystem. This array
381 * is immutable after creation apart from the init_css_set
382 * during subsystem registration (at boot time) and modular subsystem
385 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
387 /* For RCU-protected deletion */
388 struct rcu_head rcu_head
;
392 * cgroup_map_cb is an abstract callback API for reporting map-valued
396 struct cgroup_map_cb
{
397 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
402 * struct cftype: handler definitions for cgroup control files
404 * When reading/writing to a file:
405 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
406 * - the 'cftype' of the file is file->f_dentry->d_fsdata
411 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
412 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
413 CFTYPE_INSANE
= (1 << 2), /* don't create if sane_behavior */
414 CFTYPE_NO_PREFIX
= (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
417 #define MAX_CFTYPE_NAME 64
421 * By convention, the name should begin with the name of the
422 * subsystem, followed by a period. Zero length string indicates
423 * end of cftype array.
425 char name
[MAX_CFTYPE_NAME
];
428 * If not 0, file mode is set to this value, otherwise it will
429 * be figured out automatically
434 * If non-zero, defines the maximum length of string that can
435 * be passed to write_string; defaults to 64
437 size_t max_write_len
;
443 * The subsys this file belongs to. Initialized automatically
444 * during registration. NULL for cgroup core files.
446 struct cgroup_subsys
*ss
;
448 int (*open
)(struct inode
*inode
, struct file
*file
);
449 ssize_t (*read
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
451 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
453 * read_u64() is a shortcut for the common case of returning a
454 * single integer. Use it in place of read()
456 u64 (*read_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
458 * read_s64() is a signed version of read_u64()
460 s64 (*read_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
462 * read_map() is used for defining a map of key/value
463 * pairs. It should call cb->fill(cb, key, value) for each
464 * entry. The key/value pairs (and their ordering) should not
465 * change between reboots.
467 int (*read_map
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
468 struct cgroup_map_cb
*cb
);
470 * read_seq_string() is used for outputting a simple sequence
473 int (*read_seq_string
)(struct cgroup_subsys_state
*css
,
474 struct cftype
*cft
, struct seq_file
*m
);
476 ssize_t (*write
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
478 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
481 * write_u64() is a shortcut for the common case of accepting
482 * a single integer (as parsed by simple_strtoull) from
483 * userspace. Use in place of write(); return 0 or error.
485 int (*write_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
488 * write_s64() is a signed version of write_u64()
490 int (*write_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
494 * write_string() is passed a nul-terminated kernelspace
495 * buffer of maximum length determined by max_write_len.
496 * Returns 0 or -ve error code.
498 int (*write_string
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
501 * trigger() callback can be used to get some kick from the
502 * userspace, when the actual string written is not important
503 * at all. The private field can be used to determine the
504 * kick type for multiplexing.
506 int (*trigger
)(struct cgroup_subsys_state
*css
, unsigned int event
);
508 int (*release
)(struct inode
*inode
, struct file
*file
);
511 * register_event() callback will be used to add new userspace
512 * waiter for changes related to the cftype. Implement it if
513 * you want to provide this functionality. Use eventfd_signal()
514 * on eventfd to send notification to userspace.
516 int (*register_event
)(struct cgroup_subsys_state
*css
,
517 struct cftype
*cft
, struct eventfd_ctx
*eventfd
,
520 * unregister_event() callback will be called when userspace
521 * closes the eventfd or on cgroup removing.
522 * This callback must be implemented, if you want provide
523 * notification functionality.
525 void (*unregister_event
)(struct cgroup_subsys_state
*css
,
527 struct eventfd_ctx
*eventfd
);
531 * cftype_sets describe cftypes belonging to a subsystem and are chained at
532 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
533 * terminated by zero length name.
536 struct list_head node
; /* chained at subsys->cftsets */
541 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
542 * function can be called as long as @cgrp is accessible.
544 static inline bool cgroup_sane_behavior(const struct cgroup
*cgrp
)
546 return cgrp
->root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
;
549 /* Caller should hold rcu_read_lock() */
550 static inline const char *cgroup_name(const struct cgroup
*cgrp
)
552 return rcu_dereference(cgrp
->name
)->name
;
555 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
556 int cgroup_rm_cftypes(struct cftype
*cfts
);
558 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
);
560 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
561 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
);
563 int cgroup_task_count(const struct cgroup
*cgrp
);
566 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
569 struct cgroup_taskset
;
570 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
571 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
572 struct cgroup_subsys_state
*cgroup_taskset_cur_css(struct cgroup_taskset
*tset
,
574 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
577 * cgroup_taskset_for_each - iterate cgroup_taskset
578 * @task: the loop cursor
579 * @skip_css: skip if task's css matches this, %NULL to iterate through all
580 * @tset: taskset to iterate
582 #define cgroup_taskset_for_each(task, skip_css, tset) \
583 for ((task) = cgroup_taskset_first((tset)); (task); \
584 (task) = cgroup_taskset_next((tset))) \
586 cgroup_taskset_cur_css((tset), \
587 (skip_css)->ss->subsys_id) != (skip_css))
590 * Control Group subsystem type.
591 * See Documentation/cgroups/cgroups.txt for details
594 struct cgroup_subsys
{
595 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup_subsys_state
*parent_css
);
596 int (*css_online
)(struct cgroup_subsys_state
*css
);
597 void (*css_offline
)(struct cgroup_subsys_state
*css
);
598 void (*css_free
)(struct cgroup_subsys_state
*css
);
600 int (*can_attach
)(struct cgroup_subsys_state
*css
,
601 struct cgroup_taskset
*tset
);
602 void (*cancel_attach
)(struct cgroup_subsys_state
*css
,
603 struct cgroup_taskset
*tset
);
604 void (*attach
)(struct cgroup_subsys_state
*css
,
605 struct cgroup_taskset
*tset
);
606 void (*fork
)(struct task_struct
*task
);
607 void (*exit
)(struct cgroup_subsys_state
*css
,
608 struct cgroup_subsys_state
*old_css
,
609 struct task_struct
*task
);
610 void (*bind
)(struct cgroup_subsys_state
*root_css
);
616 * True if this subsys uses ID. ID is not available before cgroup_init()
617 * (not available in early_init time.)
622 * If %false, this subsystem is properly hierarchical -
623 * configuration, resource accounting and restriction on a parent
624 * cgroup cover those of its children. If %true, hierarchy support
625 * is broken in some ways - some subsystems ignore hierarchy
626 * completely while others are only implemented half-way.
628 * It's now disallowed to create nested cgroups if the subsystem is
629 * broken and cgroup core will emit a warning message on such
630 * cases. Eventually, all subsystems will be made properly
631 * hierarchical and this will go away.
633 bool broken_hierarchy
;
634 bool warned_broken_hierarchy
;
636 #define MAX_CGROUP_TYPE_NAMELEN 32
640 * Link to parent, and list entry in parent's children.
641 * Protected by cgroup_lock()
643 struct cgroupfs_root
*root
;
644 struct list_head sibling
;
645 /* used when use_id == true */
649 /* list of cftype_sets */
650 struct list_head cftsets
;
652 /* base cftypes, automatically [de]registered with subsys itself */
653 struct cftype
*base_cftypes
;
654 struct cftype_set base_cftset
;
656 /* should be defined only by modular subsystems */
657 struct module
*module
;
660 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
661 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
662 #include <linux/cgroup_subsys.h>
663 #undef IS_SUBSYS_ENABLED
667 * css_parent - find the parent css
668 * @css: the target cgroup_subsys_state
670 * Return the parent css of @css. This function is guaranteed to return
671 * non-NULL parent as long as @css isn't the root.
674 struct cgroup_subsys_state
*css_parent(struct cgroup_subsys_state
*css
)
680 * task_css_set_check - obtain a task's css_set with extra access conditions
681 * @task: the task to obtain css_set for
682 * @__c: extra condition expression to be passed to rcu_dereference_check()
684 * A task's css_set is RCU protected, initialized and exited while holding
685 * task_lock(), and can only be modified while holding both cgroup_mutex
686 * and task_lock() while the task is alive. This macro verifies that the
687 * caller is inside proper critical section and returns @task's css_set.
689 * The caller can also specify additional allowed conditions via @__c, such
690 * as locks used during the cgroup_subsys::attach() methods.
692 #ifdef CONFIG_PROVE_RCU
693 extern struct mutex cgroup_mutex
;
694 #define task_css_set_check(task, __c) \
695 rcu_dereference_check((task)->cgroups, \
696 lockdep_is_held(&(task)->alloc_lock) || \
697 lockdep_is_held(&cgroup_mutex) || (__c))
699 #define task_css_set_check(task, __c) \
700 rcu_dereference((task)->cgroups)
704 * task_css_check - obtain css for (task, subsys) w/ extra access conds
705 * @task: the target task
706 * @subsys_id: the target subsystem ID
707 * @__c: extra condition expression to be passed to rcu_dereference_check()
709 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
710 * synchronization rules are the same as task_css_set_check().
712 #define task_css_check(task, subsys_id, __c) \
713 task_css_set_check((task), (__c))->subsys[(subsys_id)]
716 * task_css_set - obtain a task's css_set
717 * @task: the task to obtain css_set for
719 * See task_css_set_check().
721 static inline struct css_set
*task_css_set(struct task_struct
*task
)
723 return task_css_set_check(task
, false);
727 * task_css - obtain css for (task, subsys)
728 * @task: the target task
729 * @subsys_id: the target subsystem ID
731 * See task_css_check().
733 static inline struct cgroup_subsys_state
*task_css(struct task_struct
*task
,
736 return task_css_check(task
, subsys_id
, false);
739 static inline struct cgroup
*task_cgroup(struct task_struct
*task
,
742 return task_css(task
, subsys_id
)->cgroup
;
745 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
746 struct cgroup_subsys_state
*parent
);
748 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
);
751 * css_for_each_child - iterate through children of a css
752 * @pos: the css * to use as the loop cursor
753 * @parent: css whose children to walk
755 * Walk @parent's children. Must be called under rcu_read_lock(). A child
756 * css which hasn't finished ->css_online() or already has finished
757 * ->css_offline() may show up during traversal and it's each subsystem's
758 * responsibility to verify that each @pos is alive.
760 * If a subsystem synchronizes against the parent in its ->css_online() and
761 * before starting iterating, a css which finished ->css_online() is
762 * guaranteed to be visible in the future iterations.
764 * It is allowed to temporarily drop RCU read lock during iteration. The
765 * caller is responsible for ensuring that @pos remains accessible until
766 * the start of the next iteration by, for example, bumping the css refcnt.
768 #define css_for_each_child(pos, parent) \
769 for ((pos) = css_next_child(NULL, (parent)); (pos); \
770 (pos) = css_next_child((pos), (parent)))
772 struct cgroup_subsys_state
*
773 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
774 struct cgroup_subsys_state
*css
);
776 struct cgroup_subsys_state
*
777 css_rightmost_descendant(struct cgroup_subsys_state
*pos
);
780 * css_for_each_descendant_pre - pre-order walk of a css's descendants
781 * @pos: the css * to use as the loop cursor
782 * @root: css whose descendants to walk
784 * Walk @root's descendants. @root is included in the iteration and the
785 * first node to be visited. Must be called under rcu_read_lock(). A
786 * descendant css which hasn't finished ->css_online() or already has
787 * finished ->css_offline() may show up during traversal and it's each
788 * subsystem's responsibility to verify that each @pos is alive.
790 * If a subsystem synchronizes against the parent in its ->css_online() and
791 * before starting iterating, and synchronizes against @pos on each
792 * iteration, any descendant css which finished ->css_online() is
793 * guaranteed to be visible in the future iterations.
795 * In other words, the following guarantees that a descendant can't escape
796 * state updates of its ancestors.
800 * Lock @css's parent and @css;
801 * Inherit state from the parent;
805 * my_update_state(@css)
807 * css_for_each_descendant_pre(@pos, @css) {
810 * Update @css's state;
812 * Verify @pos is alive and inherit state from its parent;
817 * As long as the inheriting step, including checking the parent state, is
818 * enclosed inside @pos locking, double-locking the parent isn't necessary
819 * while inheriting. The state update to the parent is guaranteed to be
820 * visible by walking order and, as long as inheriting operations to the
821 * same @pos are atomic to each other, multiple updates racing each other
822 * still result in the correct state. It's guaranateed that at least one
823 * inheritance happens for any css after the latest update to its parent.
825 * If checking parent's state requires locking the parent, each inheriting
826 * iteration should lock and unlock both @pos->parent and @pos.
828 * Alternatively, a subsystem may choose to use a single global lock to
829 * synchronize ->css_online() and ->css_offline() against tree-walking
832 * It is allowed to temporarily drop RCU read lock during iteration. The
833 * caller is responsible for ensuring that @pos remains accessible until
834 * the start of the next iteration by, for example, bumping the css refcnt.
836 #define css_for_each_descendant_pre(pos, css) \
837 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
838 (pos) = css_next_descendant_pre((pos), (css)))
840 struct cgroup_subsys_state
*
841 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
842 struct cgroup_subsys_state
*css
);
845 * css_for_each_descendant_post - post-order walk of a css's descendants
846 * @pos: the css * to use as the loop cursor
847 * @css: css whose descendants to walk
849 * Similar to css_for_each_descendant_pre() but performs post-order
850 * traversal instead. @root is included in the iteration and the last
851 * node to be visited. Note that the walk visibility guarantee described
852 * in pre-order walk doesn't apply the same to post-order walks.
854 #define css_for_each_descendant_post(pos, css) \
855 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
856 (pos) = css_next_descendant_post((pos), (css)))
858 /* A css_task_iter should be treated as an opaque object */
859 struct css_task_iter
{
860 struct cgroup_subsys_state
*origin_css
;
861 struct list_head
*cset_link
;
862 struct list_head
*task
;
865 void css_task_iter_start(struct cgroup_subsys_state
*css
,
866 struct css_task_iter
*it
);
867 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
);
868 void css_task_iter_end(struct css_task_iter
*it
);
870 int css_scan_tasks(struct cgroup_subsys_state
*css
,
871 bool (*test
)(struct task_struct
*, void *),
872 void (*process
)(struct task_struct
*, void *),
873 void *data
, struct ptr_heap
*heap
);
875 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
876 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
);
879 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
880 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
881 * CSS ID is assigned at cgroup allocation (create) automatically
882 * and removed when subsys calls free_css_id() function. This is because
883 * the lifetime of cgroup_subsys_state is subsys's matter.
885 * Looking up and scanning function should be called under rcu_read_lock().
886 * Taking cgroup_mutex is not necessary for following calls.
887 * But the css returned by this routine can be "not populated yet" or "being
888 * destroyed". The caller should check css and cgroup's status.
892 * Typically Called at ->destroy(), or somewhere the subsys frees
893 * cgroup_subsys_state.
895 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
897 /* Find a cgroup_subsys_state which has given ID */
899 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
901 /* Returns true if root is ancestor of cg */
902 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
903 const struct cgroup_subsys_state
*root
);
905 /* Get id and depth of css */
906 unsigned short css_id(struct cgroup_subsys_state
*css
);
907 struct cgroup_subsys_state
*css_from_dir(struct dentry
*dentry
,
908 struct cgroup_subsys
*ss
);
910 #else /* !CONFIG_CGROUPS */
912 static inline int cgroup_init_early(void) { return 0; }
913 static inline int cgroup_init(void) { return 0; }
914 static inline void cgroup_fork(struct task_struct
*p
) {}
915 static inline void cgroup_post_fork(struct task_struct
*p
) {}
916 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
918 static inline int cgroupstats_build(struct cgroupstats
*stats
,
919 struct dentry
*dentry
)
924 /* No cgroups - nothing to do */
925 static inline int cgroup_attach_task_all(struct task_struct
*from
,
926 struct task_struct
*t
)
931 #endif /* !CONFIG_CGROUPS */
933 #endif /* _LINUX_CGROUP_H */