1 #ifndef _LINUX_CGROUP_H
2 #define _LINUX_CGROUP_H
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
11 #include <linux/sched.h>
12 #include <linux/cpumask.h>
13 #include <linux/nodemask.h>
14 #include <linux/rcupdate.h>
15 #include <linux/rculist.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/rwsem.h>
18 #include <linux/idr.h>
19 #include <linux/workqueue.h>
21 #include <linux/percpu-refcount.h>
22 #include <linux/seq_file.h>
23 #include <linux/kernfs.h>
24 #include <linux/wait.h>
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 *m, struct pid_namespace *ns,
41 struct pid *pid, struct task_struct *tsk);
43 /* define the enumeration of all cgroup subsystems */
44 #define SUBSYS(_x) _x ## _cgrp_id,
45 enum cgroup_subsys_id {
46 #include <linux/cgroup_subsys.h>
52 * Per-subsystem/per-cgroup state maintained by the system. This is the
53 * fundamental structural building block that controllers deal with.
55 * Fields marked with "PI:" are public and immutable and may be accessed
56 * directly without synchronization.
58 struct cgroup_subsys_state {
59 /* PI: the cgroup that this css is attached to */
60 struct cgroup *cgroup;
62 /* PI: the cgroup subsystem that this css is attached to */
63 struct cgroup_subsys *ss;
65 /* reference count - access via css_[try]get() and css_put() */
66 struct percpu_ref refcnt;
68 /* PI: the parent css */
69 struct cgroup_subsys_state *parent;
71 /* siblings list anchored at the parent's ->children */
72 struct list_head sibling;
73 struct list_head children;
76 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
77 * matching css can be looked up using css_from_id().
84 * Monotonically increasing unique serial number which defines a
85 * uniform order among all csses. It's guaranteed that all
86 * ->children lists are in the ascending order of ->serial_nr and
87 * used to allow interrupting and resuming iterations.
91 /* percpu_ref killing and RCU release */
92 struct rcu_head rcu_head;
93 struct work_struct destroy_work;
96 /* bits in struct cgroup_subsys_state flags field */
98 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
99 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
100 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
104 * css_get - obtain a reference on the specified css
107 * The caller must already have a reference.
109 static inline void css_get(struct cgroup_subsys_state *css)
111 if (!(css->flags & CSS_NO_REF))
112 percpu_ref_get(&css->refcnt);
116 * css_tryget - try to obtain a reference on the specified css
119 * Obtain a reference on @css unless it already has reached zero and is
120 * being released. This function doesn't care whether @css is on or
121 * offline. The caller naturally needs to ensure that @css is accessible
122 * but doesn't have to be holding a reference on it - IOW, RCU protected
123 * access is good enough for this function. Returns %true if a reference
124 * count was successfully obtained; %false otherwise.
126 static inline bool css_tryget(struct cgroup_subsys_state *css)
128 if (!(css->flags & CSS_NO_REF))
129 return percpu_ref_tryget(&css->refcnt);
134 * css_tryget_online - try to obtain a reference on the specified css if online
137 * Obtain a reference on @css if it's online. The caller naturally needs
138 * to ensure that @css is accessible but doesn't have to be holding a
139 * reference on it - IOW, RCU protected access is good enough for this
140 * function. Returns %true if a reference count was successfully obtained;
143 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
145 if (!(css->flags & CSS_NO_REF))
146 return percpu_ref_tryget_live(&css->refcnt);
151 * css_put - put a css reference
154 * Put a reference obtained via css_get() and css_tryget_online().
156 static inline void css_put(struct cgroup_subsys_state *css)
158 if (!(css->flags & CSS_NO_REF))
159 percpu_ref_put(&css->refcnt);
162 /* bits in struct cgroup flags field */
165 * Control Group has previously had a child cgroup or a task,
166 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
169 /* Control Group requires release notifications to userspace */
170 CGRP_NOTIFY_ON_RELEASE,
172 * Clone the parent's configuration when creating a new child
173 * cpuset cgroup. For historical reasons, this option can be
174 * specified at mount time and thus is implemented here.
176 CGRP_CPUSET_CLONE_CHILDREN,
180 /* self css with NULL ->ss, points back to this cgroup */
181 struct cgroup_subsys_state self;
183 unsigned long flags; /* "unsigned long" so bitops work */
186 * idr allocated in-hierarchy ID.
188 * ID 0 is not used, the ID of the root cgroup is always 1, and a
189 * new cgroup will be assigned with a smallest available ID.
191 * Allocating/Removing ID must be protected by cgroup_mutex.
196 * If this cgroup contains any tasks, it contributes one to
197 * populated_cnt. All children with non-zero popuplated_cnt of
198 * their own contribute one. The count is zero iff there's no task
199 * in this cgroup or its subtree.
203 struct kernfs_node *kn; /* cgroup kernfs entry */
204 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
207 * The bitmask of subsystems enabled on the child cgroups.
208 * ->subtree_control is the one configured through
209 * "cgroup.subtree_control" while ->child_subsys_mask is the
210 * effective one which may have more subsystems enabled.
211 * Controller knobs are made available iff it's enabled in
214 unsigned int subtree_control;
215 unsigned int child_subsys_mask;
217 /* Private pointers for each registered subsystem */
218 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
220 struct cgroup_root *root;
223 * List of cgrp_cset_links pointing at css_sets with tasks in this
224 * cgroup. Protected by css_set_lock.
226 struct list_head cset_links;
229 * On the default hierarchy, a css_set for a cgroup with some
230 * susbsys disabled will point to css's which are associated with
231 * the closest ancestor which has the subsys enabled. The
232 * following lists all css_sets which point to this cgroup's css
233 * for the given subsystem.
235 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
238 * list of pidlists, up to two for each namespace (one for procs, one
239 * for tasks); created on demand.
241 struct list_head pidlists;
242 struct mutex pidlist_mutex;
244 /* used to wait for offlining of csses */
245 wait_queue_head_t offline_waitq;
247 /* used to schedule release agent */
248 struct work_struct release_agent_work;
251 #define MAX_CGROUP_ROOT_NAMELEN 64
253 /* cgroup_root->flags */
255 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
256 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
257 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
261 * A cgroup_root represents the root of a cgroup hierarchy, and may be
262 * associated with a kernfs_root to form an active hierarchy. This is
263 * internal to cgroup core. Don't access directly from controllers.
266 struct kernfs_root *kf_root;
268 /* The bitmask of subsystems attached to this hierarchy */
269 unsigned int subsys_mask;
271 /* Unique id for this hierarchy. */
274 /* The root cgroup. Root is destroyed on its release. */
277 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
280 /* A list running through the active hierarchies */
281 struct list_head root_list;
283 /* Hierarchy-specific flags */
286 /* IDs for cgroups in this hierarchy */
287 struct idr cgroup_idr;
289 /* The path to use for release notifications. */
290 char release_agent_path[PATH_MAX];
292 /* The name for this hierarchy - may be empty */
293 char name[MAX_CGROUP_ROOT_NAMELEN];
297 * A css_set is a structure holding pointers to a set of
298 * cgroup_subsys_state objects. This saves space in the task struct
299 * object and speeds up fork()/exit(), since a single inc/dec and a
300 * list_add()/del() can bump the reference count on the entire cgroup
306 /* Reference count */
310 * List running through all cgroup groups in the same hash
311 * slot. Protected by css_set_lock
313 struct hlist_node hlist;
316 * Lists running through all tasks using this cgroup group.
317 * mg_tasks lists tasks which belong to this cset but are in the
318 * process of being migrated out or in. Protected by
319 * css_set_rwsem, but, during migration, once tasks are moved to
320 * mg_tasks, it can be read safely while holding cgroup_mutex.
322 struct list_head tasks;
323 struct list_head mg_tasks;
326 * List of cgrp_cset_links pointing at cgroups referenced from this
327 * css_set. Protected by css_set_lock.
329 struct list_head cgrp_links;
331 /* the default cgroup associated with this css_set */
332 struct cgroup *dfl_cgrp;
335 * Set of subsystem states, one for each subsystem. This array is
336 * immutable after creation apart from the init_css_set during
337 * subsystem registration (at boot time).
339 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
342 * List of csets participating in the on-going migration either as
343 * source or destination. Protected by cgroup_mutex.
345 struct list_head mg_preload_node;
346 struct list_head mg_node;
349 * If this cset is acting as the source of migration the following
350 * two fields are set. mg_src_cgrp is the source cgroup of the
351 * on-going migration and mg_dst_cset is the destination cset the
352 * target tasks on this cset should be migrated to. Protected by
355 struct cgroup *mg_src_cgrp;
356 struct css_set *mg_dst_cset;
359 * On the default hierarhcy, ->subsys[ssid] may point to a css
360 * attached to an ancestor instead of the cgroup this css_set is
361 * associated with. The following node is anchored at
362 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
363 * iterate through all css's attached to a given cgroup.
365 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
367 /* For RCU-protected deletion */
368 struct rcu_head rcu_head;
372 * struct cftype: handler definitions for cgroup control files
374 * When reading/writing to a file:
375 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
376 * - the 'cftype' of the file is file->f_dentry->d_fsdata
381 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
382 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
383 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
385 /* internal flags, do not use outside cgroup core proper */
386 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
387 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
390 #define MAX_CFTYPE_NAME 64
394 * By convention, the name should begin with the name of the
395 * subsystem, followed by a period. Zero length string indicates
396 * end of cftype array.
398 char name[MAX_CFTYPE_NAME];
401 * If not 0, file mode is set to this value, otherwise it will
402 * be figured out automatically
407 * The maximum length of string, excluding trailing nul, that can
408 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
410 size_t max_write_len;
416 * Fields used for internal bookkeeping. Initialized automatically
417 * during registration.
419 struct cgroup_subsys *ss; /* NULL for cgroup core files */
420 struct list_head node; /* anchored at ss->cfts */
421 struct kernfs_ops *kf_ops;
424 * read_u64() is a shortcut for the common case of returning a
425 * single integer. Use it in place of read()
427 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
429 * read_s64() is a signed version of read_u64()
431 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
433 /* generic seq_file read interface */
434 int (*seq_show)(struct seq_file *sf, void *v);
436 /* optional ops, implement all or none */
437 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
438 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
439 void (*seq_stop)(struct seq_file *sf, void *v);
442 * write_u64() is a shortcut for the common case of accepting
443 * a single integer (as parsed by simple_strtoull) from
444 * userspace. Use in place of write(); return 0 or error.
446 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
449 * write_s64() is a signed version of write_u64()
451 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
455 * write() is the generic write callback which maps directly to
456 * kernfs write operation and overrides all other operations.
457 * Maximum write size is determined by ->max_write_len. Use
458 * of_css/cft() to access the associated css and cft.
460 ssize_t (*write)(struct kernfs_open_file *of,
461 char *buf, size_t nbytes, loff_t off);
463 #ifdef CONFIG_DEBUG_LOCK_ALLOC
464 struct lock_class_key lockdep_key;
468 extern struct cgroup_root cgrp_dfl_root;
469 extern struct css_set init_css_set;
472 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
473 * @cgrp: the cgroup of interest
475 * The default hierarchy is the v2 interface of cgroup and this function
476 * can be used to test whether a cgroup is on the default hierarchy for
477 * cases where a subsystem should behave differnetly depending on the
480 * The set of behaviors which change on the default hierarchy are still
481 * being determined and the mount option is prefixed with __DEVEL__.
483 * List of changed behaviors:
485 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
486 * and "name" are disallowed.
488 * - When mounting an existing superblock, mount options should match.
490 * - Remount is disallowed.
492 * - rename(2) is disallowed.
494 * - "tasks" is removed. Everything should be at process granularity. Use
495 * "cgroup.procs" instead.
497 * - "cgroup.procs" is not sorted. pids will be unique unless they got
498 * recycled inbetween reads.
500 * - "release_agent" and "notify_on_release" are removed. Replacement
501 * notification mechanism will be implemented.
503 * - "cgroup.clone_children" is removed.
505 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
506 * and its descendants contain no task; otherwise, 1. The file also
507 * generates kernfs notification which can be monitored through poll and
508 * [di]notify when the value of the file changes.
510 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
511 * take masks of ancestors with non-empty cpus/mems, instead of being
512 * moved to an ancestor.
514 * - cpuset: a task can be moved into an empty cpuset, and again it takes
515 * masks of ancestors.
517 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
520 * - blkcg: blk-throttle becomes properly hierarchical.
522 * - debug: disallowed on the default hierarchy.
524 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
526 return cgrp->root == &cgrp_dfl_root;
529 /* no synchronization, the result can only be used as a hint */
530 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
532 return !list_empty(&cgrp->cset_links);
535 /* returns ino associated with a cgroup */
536 static inline ino_t cgroup_ino(struct cgroup *cgrp)
538 return cgrp->kn->ino;
541 /* cft/css accessors for cftype->write() operation */
542 static inline struct cftype *of_cft(struct kernfs_open_file *of)
547 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
549 /* cft/css accessors for cftype->seq_*() operations */
550 static inline struct cftype *seq_cft(struct seq_file *seq)
552 return of_cft(seq->private);
555 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
557 return of_css(seq->private);
561 * Name / path handling functions. All are thin wrappers around the kernfs
562 * counterparts and can be called under any context.
565 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
567 return kernfs_name(cgrp->kn, buf, buflen);
570 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
573 return kernfs_path(cgrp->kn, buf, buflen);
576 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
578 pr_cont_kernfs_name(cgrp->kn);
581 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
583 pr_cont_kernfs_path(cgrp->kn);
586 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
588 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
589 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
590 int cgroup_rm_cftypes(struct cftype *cfts);
592 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
595 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
598 struct cgroup_taskset;
599 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
600 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
603 * cgroup_taskset_for_each - iterate cgroup_taskset
604 * @task: the loop cursor
605 * @tset: taskset to iterate
607 #define cgroup_taskset_for_each(task, tset) \
608 for ((task) = cgroup_taskset_first((tset)); (task); \
609 (task) = cgroup_taskset_next((tset)))
612 * Control Group subsystem type.
613 * See Documentation/cgroups/cgroups.txt for details
616 struct cgroup_subsys {
617 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
618 int (*css_online)(struct cgroup_subsys_state *css);
619 void (*css_offline)(struct cgroup_subsys_state *css);
620 void (*css_free)(struct cgroup_subsys_state *css);
621 void (*css_reset)(struct cgroup_subsys_state *css);
623 int (*can_attach)(struct cgroup_subsys_state *css,
624 struct cgroup_taskset *tset);
625 void (*cancel_attach)(struct cgroup_subsys_state *css,
626 struct cgroup_taskset *tset);
627 void (*attach)(struct cgroup_subsys_state *css,
628 struct cgroup_taskset *tset);
629 void (*fork)(struct task_struct *task);
630 void (*exit)(struct cgroup_subsys_state *css,
631 struct cgroup_subsys_state *old_css,
632 struct task_struct *task);
633 void (*bind)(struct cgroup_subsys_state *root_css);
639 * If %false, this subsystem is properly hierarchical -
640 * configuration, resource accounting and restriction on a parent
641 * cgroup cover those of its children. If %true, hierarchy support
642 * is broken in some ways - some subsystems ignore hierarchy
643 * completely while others are only implemented half-way.
645 * It's now disallowed to create nested cgroups if the subsystem is
646 * broken and cgroup core will emit a warning message on such
647 * cases. Eventually, all subsystems will be made properly
648 * hierarchical and this will go away.
650 bool broken_hierarchy;
651 bool warned_broken_hierarchy;
653 /* the following two fields are initialized automtically during boot */
655 #define MAX_CGROUP_TYPE_NAMELEN 32
658 /* link to parent, protected by cgroup_lock() */
659 struct cgroup_root *root;
661 /* idr for css->id */
665 * List of cftypes. Each entry is the first entry of an array
666 * terminated by zero length name.
668 struct list_head cfts;
671 * Base cftypes which are automatically registered. The two can
672 * point to the same array.
674 struct cftype *dfl_cftypes; /* for the default hierarchy */
675 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
678 * A subsystem may depend on other subsystems. When such subsystem
679 * is enabled on a cgroup, the depended-upon subsystems are enabled
680 * together if available. Subsystems enabled due to dependency are
681 * not visible to userland until explicitly enabled. The following
682 * specifies the mask of subsystems that this one depends on.
684 unsigned int depends_on;
687 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
688 #include <linux/cgroup_subsys.h>
692 * task_css_set_check - obtain a task's css_set with extra access conditions
693 * @task: the task to obtain css_set for
694 * @__c: extra condition expression to be passed to rcu_dereference_check()
696 * A task's css_set is RCU protected, initialized and exited while holding
697 * task_lock(), and can only be modified while holding both cgroup_mutex
698 * and task_lock() while the task is alive. This macro verifies that the
699 * caller is inside proper critical section and returns @task's css_set.
701 * The caller can also specify additional allowed conditions via @__c, such
702 * as locks used during the cgroup_subsys::attach() methods.
704 #ifdef CONFIG_PROVE_RCU
705 extern struct mutex cgroup_mutex;
706 extern struct rw_semaphore css_set_rwsem;
707 #define task_css_set_check(task, __c) \
708 rcu_dereference_check((task)->cgroups, \
709 lockdep_is_held(&cgroup_mutex) || \
710 lockdep_is_held(&css_set_rwsem) || \
711 ((task)->flags & PF_EXITING) || (__c))
713 #define task_css_set_check(task, __c) \
714 rcu_dereference((task)->cgroups)
718 * task_css_check - obtain css for (task, subsys) w/ extra access conds
719 * @task: the target task
720 * @subsys_id: the target subsystem ID
721 * @__c: extra condition expression to be passed to rcu_dereference_check()
723 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
724 * synchronization rules are the same as task_css_set_check().
726 #define task_css_check(task, subsys_id, __c) \
727 task_css_set_check((task), (__c))->subsys[(subsys_id)]
730 * task_css_set - obtain a task's css_set
731 * @task: the task to obtain css_set for
733 * See task_css_set_check().
735 static inline struct css_set *task_css_set(struct task_struct *task)
737 return task_css_set_check(task, false);
741 * task_css - obtain css for (task, subsys)
742 * @task: the target task
743 * @subsys_id: the target subsystem ID
745 * See task_css_check().
747 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
750 return task_css_check(task, subsys_id, false);
754 * task_css_is_root - test whether a task belongs to the root css
755 * @task: the target task
756 * @subsys_id: the target subsystem ID
758 * Test whether @task belongs to the root css on the specified subsystem.
759 * May be invoked in any context.
761 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
763 return task_css_check(task, subsys_id, true) ==
764 init_css_set.subsys[subsys_id];
767 static inline struct cgroup *task_cgroup(struct task_struct *task,
770 return task_css(task, subsys_id)->cgroup;
773 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
774 struct cgroup_subsys_state *parent);
776 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
779 * css_for_each_child - iterate through children of a css
780 * @pos: the css * to use as the loop cursor
781 * @parent: css whose children to walk
783 * Walk @parent's children. Must be called under rcu_read_lock().
785 * If a subsystem synchronizes ->css_online() and the start of iteration, a
786 * css which finished ->css_online() is guaranteed to be visible in the
787 * future iterations and will stay visible until the last reference is put.
788 * A css which hasn't finished ->css_online() or already finished
789 * ->css_offline() may show up during traversal. It's each subsystem's
790 * responsibility to synchronize against on/offlining.
792 * It is allowed to temporarily drop RCU read lock during iteration. The
793 * caller is responsible for ensuring that @pos remains accessible until
794 * the start of the next iteration by, for example, bumping the css refcnt.
796 #define css_for_each_child(pos, parent) \
797 for ((pos) = css_next_child(NULL, (parent)); (pos); \
798 (pos) = css_next_child((pos), (parent)))
800 struct cgroup_subsys_state *
801 css_next_descendant_pre(struct cgroup_subsys_state *pos,
802 struct cgroup_subsys_state *css);
804 struct cgroup_subsys_state *
805 css_rightmost_descendant(struct cgroup_subsys_state *pos);
808 * css_for_each_descendant_pre - pre-order walk of a css's descendants
809 * @pos: the css * to use as the loop cursor
810 * @root: css whose descendants to walk
812 * Walk @root's descendants. @root is included in the iteration and the
813 * first node to be visited. Must be called under rcu_read_lock().
815 * If a subsystem synchronizes ->css_online() and the start of iteration, a
816 * css which finished ->css_online() is guaranteed to be visible in the
817 * future iterations and will stay visible until the last reference is put.
818 * A css which hasn't finished ->css_online() or already finished
819 * ->css_offline() may show up during traversal. It's each subsystem's
820 * responsibility to synchronize against on/offlining.
822 * For example, the following guarantees that a descendant can't escape
823 * state updates of its ancestors.
827 * Lock @css's parent and @css;
828 * Inherit state from the parent;
832 * my_update_state(@css)
834 * css_for_each_descendant_pre(@pos, @css) {
837 * Update @css's state;
839 * Verify @pos is alive and inherit state from its parent;
844 * As long as the inheriting step, including checking the parent state, is
845 * enclosed inside @pos locking, double-locking the parent isn't necessary
846 * while inheriting. The state update to the parent is guaranteed to be
847 * visible by walking order and, as long as inheriting operations to the
848 * same @pos are atomic to each other, multiple updates racing each other
849 * still result in the correct state. It's guaranateed that at least one
850 * inheritance happens for any css after the latest update to its parent.
852 * If checking parent's state requires locking the parent, each inheriting
853 * iteration should lock and unlock both @pos->parent and @pos.
855 * Alternatively, a subsystem may choose to use a single global lock to
856 * synchronize ->css_online() and ->css_offline() against tree-walking
859 * It is allowed to temporarily drop RCU read lock during iteration. The
860 * caller is responsible for ensuring that @pos remains accessible until
861 * the start of the next iteration by, for example, bumping the css refcnt.
863 #define css_for_each_descendant_pre(pos, css) \
864 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
865 (pos) = css_next_descendant_pre((pos), (css)))
867 struct cgroup_subsys_state *
868 css_next_descendant_post(struct cgroup_subsys_state *pos,
869 struct cgroup_subsys_state *css);
872 * css_for_each_descendant_post - post-order walk of a css's descendants
873 * @pos: the css * to use as the loop cursor
874 * @css: css whose descendants to walk
876 * Similar to css_for_each_descendant_pre() but performs post-order
877 * traversal instead. @root is included in the iteration and the last
878 * node to be visited.
880 * If a subsystem synchronizes ->css_online() and the start of iteration, a
881 * css which finished ->css_online() is guaranteed to be visible in the
882 * future iterations and will stay visible until the last reference is put.
883 * A css which hasn't finished ->css_online() or already finished
884 * ->css_offline() may show up during traversal. It's each subsystem's
885 * responsibility to synchronize against on/offlining.
887 * Note that the walk visibility guarantee example described in pre-order
888 * walk doesn't apply the same to post-order walks.
890 #define css_for_each_descendant_post(pos, css) \
891 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
892 (pos) = css_next_descendant_post((pos), (css)))
894 bool css_has_online_children(struct cgroup_subsys_state *css);
896 /* A css_task_iter should be treated as an opaque object */
897 struct css_task_iter {
898 struct cgroup_subsys *ss;
900 struct list_head *cset_pos;
901 struct list_head *cset_head;
903 struct list_head *task_pos;
904 struct list_head *tasks_head;
905 struct list_head *mg_tasks_head;
908 void css_task_iter_start(struct cgroup_subsys_state *css,
909 struct css_task_iter *it);
910 struct task_struct *css_task_iter_next(struct css_task_iter *it);
911 void css_task_iter_end(struct css_task_iter *it);
913 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
914 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
916 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
917 struct cgroup_subsys *ss);
919 #else /* !CONFIG_CGROUPS */
921 static inline int cgroup_init_early(void) { return 0; }
922 static inline int cgroup_init(void) { return 0; }
923 static inline void cgroup_fork(struct task_struct *p) {}
924 static inline void cgroup_post_fork(struct task_struct *p) {}
925 static inline void cgroup_exit(struct task_struct *p) {}
927 static inline int cgroupstats_build(struct cgroupstats *stats,
928 struct dentry *dentry)
933 /* No cgroups - nothing to do */
934 static inline int cgroup_attach_task_all(struct task_struct *from,
935 struct task_struct *t)
940 #endif /* !CONFIG_CGROUPS */
942 #endif /* _LINUX_CGROUP_H */