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 /* dummy css with NULL ->ss, points back to this cgroup */
229 struct cgroup_subsys_state dummy_css
;
231 /* For css percpu_ref killing and RCU-protected deletion */
232 struct rcu_head rcu_head
;
233 struct work_struct destroy_work
;
234 atomic_t css_kill_cnt
;
236 /* List of events which userspace want to receive */
237 struct list_head event_list
;
238 spinlock_t event_list_lock
;
240 /* directory xattrs */
241 struct simple_xattrs xattrs
;
244 #define MAX_CGROUP_ROOT_NAMELEN 64
246 /* cgroupfs_root->flags */
249 * Unfortunately, cgroup core and various controllers are riddled
250 * with idiosyncrasies and pointless options. The following flag,
251 * when set, will force sane behavior - some options are forced on,
252 * others are disallowed, and some controllers will change their
253 * hierarchical or other behaviors.
255 * The set of behaviors affected by this flag are still being
256 * determined and developed and the mount option for this flag is
257 * prefixed with __DEVEL__. The prefix will be dropped once we
258 * reach the point where all behaviors are compatible with the
259 * planned unified hierarchy, which will automatically turn on this
262 * The followings are the behaviors currently affected this flag.
264 * - Mount options "noprefix" and "clone_children" are disallowed.
265 * Also, cgroupfs file cgroup.clone_children is not created.
267 * - When mounting an existing superblock, mount options should
270 * - Remount is disallowed.
272 * - rename(2) is disallowed.
274 * - "tasks" is removed. Everything should be at process
275 * granularity. Use "cgroup.procs" instead.
277 * - "release_agent" and "notify_on_release" are removed.
278 * Replacement notification mechanism will be implemented.
280 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
281 * and take masks of ancestors with non-empty cpus/mems, instead of
282 * being moved to an ancestor.
284 * - cpuset: a task can be moved into an empty cpuset, and again it
285 * takes masks of ancestors.
287 * - memcg: use_hierarchy is on by default and the cgroup file for
288 * the flag is not created.
290 * - blkcg: blk-throttle becomes properly hierarchical.
292 CGRP_ROOT_SANE_BEHAVIOR
= (1 << 0),
294 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
295 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
297 /* mount options live below bit 16 */
298 CGRP_ROOT_OPTION_MASK
= (1 << 16) - 1,
300 CGRP_ROOT_SUBSYS_BOUND
= (1 << 16), /* subsystems finished binding */
304 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
305 * associated with a superblock to form an active hierarchy. This is
306 * internal to cgroup core. Don't access directly from controllers.
308 struct cgroupfs_root
{
309 struct super_block
*sb
;
311 /* The bitmask of subsystems attached to this hierarchy */
312 unsigned long subsys_mask
;
314 /* Unique id for this hierarchy. */
317 /* A list running through the attached subsystems */
318 struct list_head subsys_list
;
320 /* The root cgroup for this hierarchy */
321 struct cgroup top_cgroup
;
323 /* Tracks how many cgroups are currently defined in hierarchy.*/
324 int number_of_cgroups
;
326 /* A list running through the active hierarchies */
327 struct list_head root_list
;
329 /* Hierarchy-specific flags */
332 /* IDs for cgroups in this hierarchy */
333 struct idr cgroup_idr
;
335 /* The path to use for release notifications. */
336 char release_agent_path
[PATH_MAX
];
338 /* The name for this hierarchy - may be empty */
339 char name
[MAX_CGROUP_ROOT_NAMELEN
];
343 * A css_set is a structure holding pointers to a set of
344 * cgroup_subsys_state objects. This saves space in the task struct
345 * object and speeds up fork()/exit(), since a single inc/dec and a
346 * list_add()/del() can bump the reference count on the entire cgroup
352 /* Reference count */
356 * List running through all cgroup groups in the same hash
357 * slot. Protected by css_set_lock
359 struct hlist_node hlist
;
362 * List running through all tasks using this cgroup
363 * group. Protected by css_set_lock
365 struct list_head 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
;
374 * Set of subsystem states, one for each subsystem. This array
375 * is immutable after creation apart from the init_css_set
376 * during subsystem registration (at boot time) and modular subsystem
379 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
381 /* For RCU-protected deletion */
382 struct rcu_head rcu_head
;
386 * cgroup_map_cb is an abstract callback API for reporting map-valued
390 struct cgroup_map_cb
{
391 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
396 * struct cftype: handler definitions for cgroup control files
398 * When reading/writing to a file:
399 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
400 * - the 'cftype' of the file is file->f_dentry->d_fsdata
405 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
406 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
407 CFTYPE_INSANE
= (1 << 2), /* don't create if sane_behavior */
410 #define MAX_CFTYPE_NAME 64
414 * By convention, the name should begin with the name of the
415 * subsystem, followed by a period. Zero length string indicates
416 * end of cftype array.
418 char name
[MAX_CFTYPE_NAME
];
421 * If not 0, file mode is set to this value, otherwise it will
422 * be figured out automatically
427 * If non-zero, defines the maximum length of string that can
428 * be passed to write_string; defaults to 64
430 size_t max_write_len
;
436 * The subsys this file belongs to. Initialized automatically
437 * during registration. NULL for cgroup core files.
439 struct cgroup_subsys
*ss
;
441 int (*open
)(struct inode
*inode
, struct file
*file
);
442 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
444 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
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
*cgrp
, struct cftype
*cft
);
451 * read_s64() is a signed version of read_u64()
453 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
455 * read_map() is used for defining a map of key/value
456 * pairs. It should call cb->fill(cb, key, value) for each
457 * entry. The key/value pairs (and their ordering) should not
458 * change between reboots.
460 int (*read_map
)(struct cgroup
*cgrp
, struct cftype
*cft
,
461 struct cgroup_map_cb
*cb
);
463 * read_seq_string() is used for outputting a simple sequence
466 int (*read_seq_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
469 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
471 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
474 * write_u64() is a shortcut for the common case of accepting
475 * a single integer (as parsed by simple_strtoull) from
476 * userspace. Use in place of write(); return 0 or error.
478 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
480 * write_s64() is a signed version of write_u64()
482 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
485 * write_string() is passed a nul-terminated kernelspace
486 * buffer of maximum length determined by max_write_len.
487 * Returns 0 or -ve error code.
489 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
492 * trigger() callback can be used to get some kick from the
493 * userspace, when the actual string written is not important
494 * at all. The private field can be used to determine the
495 * kick type for multiplexing.
497 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
499 int (*release
)(struct inode
*inode
, struct file
*file
);
502 * register_event() callback will be used to add new userspace
503 * waiter for changes related to the cftype. Implement it if
504 * you want to provide this functionality. Use eventfd_signal()
505 * on eventfd to send notification to userspace.
507 int (*register_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
508 struct eventfd_ctx
*eventfd
, const char *args
);
510 * unregister_event() callback will be called when userspace
511 * closes the eventfd or on cgroup removing.
512 * This callback must be implemented, if you want provide
513 * notification functionality.
515 void (*unregister_event
)(struct cgroup
*cgrp
, struct cftype
*cft
,
516 struct eventfd_ctx
*eventfd
);
520 * cftype_sets describe cftypes belonging to a subsystem and are chained at
521 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
522 * terminated by zero length name.
525 struct list_head node
; /* chained at subsys->cftsets */
529 struct cgroup_scanner
{
531 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
532 void (*process_task
)(struct task_struct
*p
,
533 struct cgroup_scanner
*scan
);
534 struct ptr_heap
*heap
;
539 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
540 * function can be called as long as @cgrp is accessible.
542 static inline bool cgroup_sane_behavior(const struct cgroup
*cgrp
)
544 return cgrp
->root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
;
547 /* Caller should hold rcu_read_lock() */
548 static inline const char *cgroup_name(const struct cgroup
*cgrp
)
550 return rcu_dereference(cgrp
->name
)->name
;
553 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
);
554 int cgroup_rm_cftypes(struct cftype
*cfts
);
556 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
);
558 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
559 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
);
561 int cgroup_task_count(const struct cgroup
*cgrp
);
564 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
567 struct cgroup_taskset
;
568 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
);
569 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
);
570 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
);
571 int cgroup_taskset_size(struct cgroup_taskset
*tset
);
574 * cgroup_taskset_for_each - iterate cgroup_taskset
575 * @task: the loop cursor
576 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
577 * @tset: taskset to iterate
579 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
580 for ((task) = cgroup_taskset_first((tset)); (task); \
581 (task) = cgroup_taskset_next((tset))) \
582 if (!(skip_cgrp) || \
583 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
586 * Control Group subsystem type.
587 * See Documentation/cgroups/cgroups.txt for details
590 struct cgroup_subsys
{
591 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup_subsys_state
*parent_css
);
592 int (*css_online
)(struct cgroup_subsys_state
*css
);
593 void (*css_offline
)(struct cgroup_subsys_state
*css
);
594 void (*css_free
)(struct cgroup_subsys_state
*css
);
596 int (*can_attach
)(struct cgroup_subsys_state
*css
,
597 struct cgroup_taskset
*tset
);
598 void (*cancel_attach
)(struct cgroup_subsys_state
*css
,
599 struct cgroup_taskset
*tset
);
600 void (*attach
)(struct cgroup_subsys_state
*css
,
601 struct cgroup_taskset
*tset
);
602 void (*fork
)(struct task_struct
*task
);
603 void (*exit
)(struct cgroup_subsys_state
*css
,
604 struct cgroup_subsys_state
*old_css
,
605 struct task_struct
*task
);
606 void (*bind
)(struct cgroup_subsys_state
*root_css
);
612 * True if this subsys uses ID. ID is not available before cgroup_init()
613 * (not available in early_init time.)
618 * If %false, this subsystem is properly hierarchical -
619 * configuration, resource accounting and restriction on a parent
620 * cgroup cover those of its children. If %true, hierarchy support
621 * is broken in some ways - some subsystems ignore hierarchy
622 * completely while others are only implemented half-way.
624 * It's now disallowed to create nested cgroups if the subsystem is
625 * broken and cgroup core will emit a warning message on such
626 * cases. Eventually, all subsystems will be made properly
627 * hierarchical and this will go away.
629 bool broken_hierarchy
;
630 bool warned_broken_hierarchy
;
632 #define MAX_CGROUP_TYPE_NAMELEN 32
636 * Link to parent, and list entry in parent's children.
637 * Protected by cgroup_lock()
639 struct cgroupfs_root
*root
;
640 struct list_head sibling
;
641 /* used when use_id == true */
645 /* list of cftype_sets */
646 struct list_head cftsets
;
648 /* base cftypes, automatically [de]registered with subsys itself */
649 struct cftype
*base_cftypes
;
650 struct cftype_set base_cftset
;
652 /* should be defined only by modular subsystems */
653 struct module
*module
;
656 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
657 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
658 #include <linux/cgroup_subsys.h>
659 #undef IS_SUBSYS_ENABLED
663 * css_parent - find the parent css
664 * @css: the target cgroup_subsys_state
666 * Return the parent css of @css. This function is guaranteed to return
667 * non-NULL parent as long as @css isn't the root.
670 struct cgroup_subsys_state
*css_parent(struct cgroup_subsys_state
*css
)
672 struct cgroup
*parent_cgrp
= css
->cgroup
->parent
;
678 return parent_cgrp
->subsys
[css
->ss
->subsys_id
];
680 return &parent_cgrp
->dummy_css
;
684 * cgroup_css - obtain a cgroup's css for the specified subsystem
685 * @cgrp: the cgroup of interest
686 * @subsys_id: the subsystem of interest
688 * Return @cgrp's css (cgroup_subsys_state) associated with @subsys_id.
690 static inline struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
693 return cgrp
->subsys
[subsys_id
];
697 * task_css_set_check - obtain a task's css_set with extra access conditions
698 * @task: the task to obtain css_set for
699 * @__c: extra condition expression to be passed to rcu_dereference_check()
701 * A task's css_set is RCU protected, initialized and exited while holding
702 * task_lock(), and can only be modified while holding both cgroup_mutex
703 * and task_lock() while the task is alive. This macro verifies that the
704 * caller is inside proper critical section and returns @task's css_set.
706 * The caller can also specify additional allowed conditions via @__c, such
707 * as locks used during the cgroup_subsys::attach() methods.
709 #ifdef CONFIG_PROVE_RCU
710 extern struct mutex cgroup_mutex
;
711 #define task_css_set_check(task, __c) \
712 rcu_dereference_check((task)->cgroups, \
713 lockdep_is_held(&(task)->alloc_lock) || \
714 lockdep_is_held(&cgroup_mutex) || (__c))
716 #define task_css_set_check(task, __c) \
717 rcu_dereference((task)->cgroups)
721 * task_css_check - obtain css for (task, subsys) w/ extra access conds
722 * @task: the target task
723 * @subsys_id: the target subsystem ID
724 * @__c: extra condition expression to be passed to rcu_dereference_check()
726 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
727 * synchronization rules are the same as task_css_set_check().
729 #define task_css_check(task, subsys_id, __c) \
730 task_css_set_check((task), (__c))->subsys[(subsys_id)]
733 * task_css_set - obtain a task's css_set
734 * @task: the task to obtain css_set for
736 * See task_css_set_check().
738 static inline struct css_set
*task_css_set(struct task_struct
*task
)
740 return task_css_set_check(task
, false);
744 * task_css - obtain css for (task, subsys)
745 * @task: the target task
746 * @subsys_id: the target subsystem ID
748 * See task_css_check().
750 static inline struct cgroup_subsys_state
*task_css(struct task_struct
*task
,
753 return task_css_check(task
, subsys_id
, false);
756 static inline struct cgroup
*task_cgroup(struct task_struct
*task
,
759 return task_css(task
, subsys_id
)->cgroup
;
763 * cgroup_from_id - lookup cgroup by id
764 * @ss: cgroup subsys to be looked into
767 * Returns the cgroup if there's valid one with @id, otherwise returns NULL.
768 * Should be called under rcu_read_lock().
770 static inline struct cgroup
*cgroup_from_id(struct cgroup_subsys
*ss
, int id
)
772 #ifdef CONFIG_PROVE_RCU
773 rcu_lockdep_assert(rcu_read_lock_held() ||
774 lockdep_is_held(&cgroup_mutex
),
775 "cgroup_from_id() needs proper protection");
777 return idr_find(&ss
->root
->cgroup_idr
, id
);
780 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
);
783 * cgroup_for_each_child - iterate through children of a cgroup
784 * @pos: the cgroup * to use as the loop cursor
785 * @cgrp: cgroup whose children to walk
787 * Walk @cgrp's children. Must be called under rcu_read_lock(). A child
788 * cgroup which hasn't finished ->css_online() or already has finished
789 * ->css_offline() may show up during traversal and it's each subsystem's
790 * responsibility to verify that each @pos is alive.
792 * If a subsystem synchronizes against the parent in its ->css_online() and
793 * before starting iterating, a cgroup which finished ->css_online() is
794 * guaranteed to be visible in the future iterations.
796 * It is allowed to temporarily drop RCU read lock during iteration. The
797 * caller is responsible for ensuring that @pos remains accessible until
798 * the start of the next iteration by, for example, bumping the css refcnt.
800 #define cgroup_for_each_child(pos, cgrp) \
801 for ((pos) = list_first_or_null_rcu(&(cgrp)->children, \
802 struct cgroup, sibling); \
803 (pos); (pos) = cgroup_next_sibling((pos)))
805 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
806 struct cgroup
*cgroup
);
807 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
);
810 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
811 * @pos: the cgroup * to use as the loop cursor
812 * @cgroup: cgroup whose descendants to walk
814 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
815 * descendant cgroup which hasn't finished ->css_online() or already has
816 * finished ->css_offline() may show up during traversal and it's each
817 * subsystem's responsibility to verify that each @pos is alive.
819 * If a subsystem synchronizes against the parent in its ->css_online() and
820 * before starting iterating, and synchronizes against @pos on each
821 * iteration, any descendant cgroup which finished ->css_online() is
822 * guaranteed to be visible in the future iterations.
824 * In other words, the following guarantees that a descendant can't escape
825 * state updates of its ancestors.
829 * Lock @cgrp->parent and @cgrp;
830 * Inherit state from @cgrp->parent;
834 * my_update_state(@cgrp)
837 * Update @cgrp's state;
840 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
842 * Verify @pos is alive and inherit state from @pos->parent;
847 * As long as the inheriting step, including checking the parent state, is
848 * enclosed inside @pos locking, double-locking the parent isn't necessary
849 * while inheriting. The state update to the parent is guaranteed to be
850 * visible by walking order and, as long as inheriting operations to the
851 * same @pos are atomic to each other, multiple updates racing each other
852 * still result in the correct state. It's guaranateed that at least one
853 * inheritance happens for any cgroup after the latest update to its
856 * If checking parent's state requires locking the parent, each inheriting
857 * iteration should lock and unlock both @pos->parent and @pos.
859 * Alternatively, a subsystem may choose to use a single global lock to
860 * synchronize ->css_online() and ->css_offline() against tree-walking
863 * It is allowed to temporarily drop RCU read lock during iteration. The
864 * caller is responsible for ensuring that @pos remains accessible until
865 * the start of the next iteration by, for example, bumping the css refcnt.
867 #define cgroup_for_each_descendant_pre(pos, cgroup) \
868 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
869 pos = cgroup_next_descendant_pre((pos), (cgroup)))
871 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
872 struct cgroup
*cgroup
);
875 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
876 * @pos: the cgroup * to use as the loop cursor
877 * @cgroup: cgroup whose descendants to walk
879 * Similar to cgroup_for_each_descendant_pre() but performs post-order
880 * traversal instead. Note that the walk visibility guarantee described in
881 * pre-order walk doesn't apply the same to post-order walks.
883 #define cgroup_for_each_descendant_post(pos, cgroup) \
884 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
885 pos = cgroup_next_descendant_post((pos), (cgroup)))
887 /* A cgroup_iter should be treated as an opaque object */
889 struct list_head
*cset_link
;
890 struct list_head
*task
;
894 * To iterate across the tasks in a cgroup:
896 * 1) call cgroup_iter_start to initialize an iterator
898 * 2) call cgroup_iter_next() to retrieve member tasks until it
899 * returns NULL or until you want to end the iteration
901 * 3) call cgroup_iter_end() to destroy the iterator.
903 * Or, call cgroup_scan_tasks() to iterate through every task in a
904 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
905 * the test_task() callback, but not while calling the process_task()
908 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
909 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
910 struct cgroup_iter
*it
);
911 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
912 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
913 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*);
914 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
);
917 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
918 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
919 * CSS ID is assigned at cgroup allocation (create) automatically
920 * and removed when subsys calls free_css_id() function. This is because
921 * the lifetime of cgroup_subsys_state is subsys's matter.
923 * Looking up and scanning function should be called under rcu_read_lock().
924 * Taking cgroup_mutex is not necessary for following calls.
925 * But the css returned by this routine can be "not populated yet" or "being
926 * destroyed". The caller should check css and cgroup's status.
930 * Typically Called at ->destroy(), or somewhere the subsys frees
931 * cgroup_subsys_state.
933 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
);
935 /* Find a cgroup_subsys_state which has given ID */
937 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
);
939 /* Returns true if root is ancestor of cg */
940 bool css_is_ancestor(struct cgroup_subsys_state
*cg
,
941 const struct cgroup_subsys_state
*root
);
943 /* Get id and depth of css */
944 unsigned short css_id(struct cgroup_subsys_state
*css
);
945 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
);
947 #else /* !CONFIG_CGROUPS */
949 static inline int cgroup_init_early(void) { return 0; }
950 static inline int cgroup_init(void) { return 0; }
951 static inline void cgroup_fork(struct task_struct
*p
) {}
952 static inline void cgroup_post_fork(struct task_struct
*p
) {}
953 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
955 static inline int cgroupstats_build(struct cgroupstats
*stats
,
956 struct dentry
*dentry
)
961 /* No cgroups - nothing to do */
962 static inline int cgroup_attach_task_all(struct task_struct
*from
,
963 struct task_struct
*t
)
968 #endif /* !CONFIG_CGROUPS */
970 #endif /* _LINUX_CGROUP_H */