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 {
70 * The cgroup that this subsystem is attached to. Useful
71 * for subsystems that want to know about the cgroup
74 struct cgroup *cgroup;
76 /* reference count - access via css_[try]get() and css_put() */
77 struct percpu_ref refcnt;
80 /* ID for this css, if possible */
81 struct css_id __rcu *id;
83 /* Used to put @cgroup->dentry on the last css_put() */
84 struct work_struct dput_work;
87 /* bits in struct cgroup_subsys_state flags field */
89 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
90 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
94 * css_get - obtain a reference on the specified css
97 * The caller must already have a reference.
99 static inline void css_get(struct cgroup_subsys_state *css)
101 /* We don't need to reference count the root state */
102 if (!(css->flags & CSS_ROOT))
103 percpu_ref_get(&css->refcnt);
107 * css_tryget - try to obtain a reference on the specified css
110 * Obtain a reference on @css if it's alive. The caller naturally needs to
111 * ensure that @css is accessible but doesn't have to be holding a
112 * reference on it - IOW, RCU protected access is good enough for this
113 * function. Returns %true if a reference count was successfully obtained;
116 static inline bool css_tryget(struct cgroup_subsys_state *css)
118 if (css->flags & CSS_ROOT)
120 return percpu_ref_tryget(&css->refcnt);
124 * css_put - put a css reference
127 * Put a reference obtained via css_get() and css_tryget().
129 static inline void css_put(struct cgroup_subsys_state *css)
131 if (!(css->flags & CSS_ROOT))
132 percpu_ref_put(&css->refcnt);
135 /* bits in struct cgroup flags field */
137 /* Control Group is dead */
140 * Control Group has previously had a child cgroup or a task,
141 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
144 /* Control Group requires release notifications to userspace */
145 CGRP_NOTIFY_ON_RELEASE,
147 * Clone the parent's configuration when creating a new child
148 * cpuset cgroup. For historical reasons, this option can be
149 * specified at mount time and thus is implemented here.
151 CGRP_CPUSET_CLONE_CHILDREN,
152 /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */
157 struct rcu_head rcu_head;
162 unsigned long flags; /* "unsigned long" so bitops work */
165 * idr allocated in-hierarchy ID.
167 * The ID of the root cgroup is always 0, and a new cgroup
168 * will be assigned with a smallest available ID.
173 * We link our 'sibling' struct into our parent's 'children'.
174 * Our children link their 'sibling' into our 'children'.
176 struct list_head sibling; /* my parent's children */
177 struct list_head children; /* my children */
178 struct list_head files; /* my files */
180 struct cgroup *parent; /* my parent */
181 struct dentry *dentry; /* cgroup fs entry, RCU protected */
184 * Monotonically increasing unique serial number which defines a
185 * uniform order among all cgroups. It's guaranteed that all
186 * ->children lists are in the ascending order of ->serial_nr.
187 * It's used to allow interrupting and resuming iterations.
192 * This is a copy of dentry->d_name, and it's needed because
193 * we can't use dentry->d_name in cgroup_path().
195 * You must acquire rcu_read_lock() to access cgrp->name, and
196 * the only place that can change it is rename(), which is
197 * protected by parent dir's i_mutex.
199 * Normally you should use cgroup_name() wrapper rather than
200 * access it directly.
202 struct cgroup_name __rcu *name;
204 /* Private pointers for each registered subsystem */
205 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
207 struct cgroupfs_root *root;
210 * List of cgrp_cset_links pointing at css_sets with tasks in this
211 * cgroup. Protected by css_set_lock.
213 struct list_head cset_links;
216 * Linked list running through all cgroups that can
217 * potentially be reaped by the release agent. Protected by
220 struct list_head release_list;
223 * list of pidlists, up to two for each namespace (one for procs, one
224 * for tasks); created on demand.
226 struct list_head pidlists;
227 struct mutex pidlist_mutex;
229 /* For css percpu_ref killing and RCU-protected deletion */
230 struct rcu_head rcu_head;
231 struct work_struct destroy_work;
232 atomic_t css_kill_cnt;
234 /* List of events which userspace want to receive */
235 struct list_head event_list;
236 spinlock_t event_list_lock;
238 /* directory xattrs */
239 struct simple_xattrs xattrs;
242 #define MAX_CGROUP_ROOT_NAMELEN 64
244 /* cgroupfs_root->flags */
247 * Unfortunately, cgroup core and various controllers are riddled
248 * with idiosyncrasies and pointless options. The following flag,
249 * when set, will force sane behavior - some options are forced on,
250 * others are disallowed, and some controllers will change their
251 * hierarchical or other behaviors.
253 * The set of behaviors affected by this flag are still being
254 * determined and developed and the mount option for this flag is
255 * prefixed with __DEVEL__. The prefix will be dropped once we
256 * reach the point where all behaviors are compatible with the
257 * planned unified hierarchy, which will automatically turn on this
260 * The followings are the behaviors currently affected this flag.
262 * - Mount options "noprefix" and "clone_children" are disallowed.
263 * Also, cgroupfs file cgroup.clone_children is not created.
265 * - When mounting an existing superblock, mount options should
268 * - Remount is disallowed.
270 * - rename(2) is disallowed.
272 * - "tasks" is removed. Everything should be at process
273 * granularity. Use "cgroup.procs" instead.
275 * - "release_agent" and "notify_on_release" are removed.
276 * Replacement notification mechanism will be implemented.
278 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
279 * and take masks of ancestors with non-empty cpus/mems, instead of
280 * being moved to an ancestor.
282 * - cpuset: a task can be moved into an empty cpuset, and again it
283 * takes masks of ancestors.
285 * - memcg: use_hierarchy is on by default and the cgroup file for
286 * the flag is not created.
288 * - blkcg: blk-throttle becomes properly hierarchical.
290 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
292 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
293 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
295 /* mount options live below bit 16 */
296 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
298 CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */
302 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
303 * associated with a superblock to form an active hierarchy. This is
304 * internal to cgroup core. Don't access directly from controllers.
306 struct cgroupfs_root {
307 struct super_block *sb;
309 /* The bitmask of subsystems attached to this hierarchy */
310 unsigned long subsys_mask;
312 /* Unique id for this hierarchy. */
315 /* A list running through the attached subsystems */
316 struct list_head subsys_list;
318 /* The root cgroup for this hierarchy */
319 struct cgroup top_cgroup;
321 /* Tracks how many cgroups are currently defined in hierarchy.*/
322 int number_of_cgroups;
324 /* A list running through the active hierarchies */
325 struct list_head root_list;
327 /* Hierarchy-specific flags */
330 /* IDs for cgroups in this hierarchy */
331 struct idr cgroup_idr;
333 /* The path to use for release notifications. */
334 char release_agent_path[PATH_MAX];
336 /* The name for this hierarchy - may be empty */
337 char name[MAX_CGROUP_ROOT_NAMELEN];
341 * A css_set is a structure holding pointers to a set of
342 * cgroup_subsys_state objects. This saves space in the task struct
343 * object and speeds up fork()/exit(), since a single inc/dec and a
344 * list_add()/del() can bump the reference count on the entire cgroup
350 /* Reference count */
354 * List running through all cgroup groups in the same hash
355 * slot. Protected by css_set_lock
357 struct hlist_node hlist;
360 * List running through all tasks using this cgroup
361 * group. Protected by css_set_lock
363 struct list_head tasks;
366 * List of cgrp_cset_links pointing at cgroups referenced from this
367 * css_set. Protected by css_set_lock.
369 struct list_head cgrp_links;
372 * Set of subsystem states, one for each subsystem. This array
373 * is immutable after creation apart from the init_css_set
374 * during subsystem registration (at boot time) and modular subsystem
377 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
379 /* For RCU-protected deletion */
380 struct rcu_head rcu_head;
384 * cgroup_map_cb is an abstract callback API for reporting map-valued
388 struct cgroup_map_cb {
389 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
394 * struct cftype: handler definitions for cgroup control files
396 * When reading/writing to a file:
397 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
398 * - the 'cftype' of the file is file->f_dentry->d_fsdata
403 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
404 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
405 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
408 #define MAX_CFTYPE_NAME 64
412 * By convention, the name should begin with the name of the
413 * subsystem, followed by a period. Zero length string indicates
414 * end of cftype array.
416 char name[MAX_CFTYPE_NAME];
419 * If not 0, file mode is set to this value, otherwise it will
420 * be figured out automatically
425 * If non-zero, defines the maximum length of string that can
426 * be passed to write_string; defaults to 64
428 size_t max_write_len;
433 int (*open)(struct inode *inode, struct file *file);
434 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
436 char __user *buf, size_t nbytes, loff_t *ppos);
438 * read_u64() is a shortcut for the common case of returning a
439 * single integer. Use it in place of read()
441 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
443 * read_s64() is a signed version of read_u64()
445 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
447 * read_map() is used for defining a map of key/value
448 * pairs. It should call cb->fill(cb, key, value) for each
449 * entry. The key/value pairs (and their ordering) should not
450 * change between reboots.
452 int (*read_map)(struct cgroup *cgrp, struct cftype *cft,
453 struct cgroup_map_cb *cb);
455 * read_seq_string() is used for outputting a simple sequence
458 int (*read_seq_string)(struct cgroup *cgrp, struct cftype *cft,
461 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
463 const char __user *buf, size_t nbytes, loff_t *ppos);
466 * write_u64() is a shortcut for the common case of accepting
467 * a single integer (as parsed by simple_strtoull) from
468 * userspace. Use in place of write(); return 0 or error.
470 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
472 * write_s64() is a signed version of write_u64()
474 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
477 * write_string() is passed a nul-terminated kernelspace
478 * buffer of maximum length determined by max_write_len.
479 * Returns 0 or -ve error code.
481 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
484 * trigger() callback can be used to get some kick from the
485 * userspace, when the actual string written is not important
486 * at all. The private field can be used to determine the
487 * kick type for multiplexing.
489 int (*trigger)(struct cgroup *cgrp, unsigned int event);
491 int (*release)(struct inode *inode, struct file *file);
494 * register_event() callback will be used to add new userspace
495 * waiter for changes related to the cftype. Implement it if
496 * you want to provide this functionality. Use eventfd_signal()
497 * on eventfd to send notification to userspace.
499 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
500 struct eventfd_ctx *eventfd, const char *args);
502 * unregister_event() callback will be called when userspace
503 * closes the eventfd or on cgroup removing.
504 * This callback must be implemented, if you want provide
505 * notification functionality.
507 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
508 struct eventfd_ctx *eventfd);
512 * cftype_sets describe cftypes belonging to a subsystem and are chained at
513 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
514 * terminated by zero length name.
517 struct list_head node; /* chained at subsys->cftsets */
521 struct cgroup_scanner {
523 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
524 void (*process_task)(struct task_struct *p,
525 struct cgroup_scanner *scan);
526 struct ptr_heap *heap;
531 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
532 * function can be called as long as @cgrp is accessible.
534 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
536 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
539 /* Caller should hold rcu_read_lock() */
540 static inline const char *cgroup_name(const struct cgroup *cgrp)
542 return rcu_dereference(cgrp->name)->name;
545 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
546 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
548 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
550 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
551 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
552 char *buf, size_t buflen);
554 int cgroup_task_count(const struct cgroup *cgrp);
557 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
560 struct cgroup_taskset;
561 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
562 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
563 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
564 int cgroup_taskset_size(struct cgroup_taskset *tset);
567 * cgroup_taskset_for_each - iterate cgroup_taskset
568 * @task: the loop cursor
569 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
570 * @tset: taskset to iterate
572 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
573 for ((task) = cgroup_taskset_first((tset)); (task); \
574 (task) = cgroup_taskset_next((tset))) \
575 if (!(skip_cgrp) || \
576 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
579 * Control Group subsystem type.
580 * See Documentation/cgroups/cgroups.txt for details
583 struct cgroup_subsys {
584 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
585 int (*css_online)(struct cgroup *cgrp);
586 void (*css_offline)(struct cgroup *cgrp);
587 void (*css_free)(struct cgroup *cgrp);
589 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
590 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
591 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
592 void (*fork)(struct task_struct *task);
593 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
594 struct task_struct *task);
595 void (*bind)(struct cgroup *root);
601 * True if this subsys uses ID. ID is not available before cgroup_init()
602 * (not available in early_init time.)
607 * If %false, this subsystem is properly hierarchical -
608 * configuration, resource accounting and restriction on a parent
609 * cgroup cover those of its children. If %true, hierarchy support
610 * is broken in some ways - some subsystems ignore hierarchy
611 * completely while others are only implemented half-way.
613 * It's now disallowed to create nested cgroups if the subsystem is
614 * broken and cgroup core will emit a warning message on such
615 * cases. Eventually, all subsystems will be made properly
616 * hierarchical and this will go away.
618 bool broken_hierarchy;
619 bool warned_broken_hierarchy;
621 #define MAX_CGROUP_TYPE_NAMELEN 32
625 * Link to parent, and list entry in parent's children.
626 * Protected by cgroup_lock()
628 struct cgroupfs_root *root;
629 struct list_head sibling;
630 /* used when use_id == true */
634 /* list of cftype_sets */
635 struct list_head cftsets;
637 /* base cftypes, automatically [de]registered with subsys itself */
638 struct cftype *base_cftypes;
639 struct cftype_set base_cftset;
641 /* should be defined only by modular subsystems */
642 struct module *module;
645 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
646 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
647 #include <linux/cgroup_subsys.h>
648 #undef IS_SUBSYS_ENABLED
651 static inline struct cgroup_subsys_state *cgroup_subsys_state(
652 struct cgroup *cgrp, int subsys_id)
654 return cgrp->subsys[subsys_id];
658 * task_css_set_check - obtain a task's css_set with extra access conditions
659 * @task: the task to obtain css_set for
660 * @__c: extra condition expression to be passed to rcu_dereference_check()
662 * A task's css_set is RCU protected, initialized and exited while holding
663 * task_lock(), and can only be modified while holding both cgroup_mutex
664 * and task_lock() while the task is alive. This macro verifies that the
665 * caller is inside proper critical section and returns @task's css_set.
667 * The caller can also specify additional allowed conditions via @__c, such
668 * as locks used during the cgroup_subsys::attach() methods.
670 #ifdef CONFIG_PROVE_RCU
671 extern struct mutex cgroup_mutex;
672 #define task_css_set_check(task, __c) \
673 rcu_dereference_check((task)->cgroups, \
674 lockdep_is_held(&(task)->alloc_lock) || \
675 lockdep_is_held(&cgroup_mutex) || (__c))
677 #define task_css_set_check(task, __c) \
678 rcu_dereference((task)->cgroups)
682 * task_subsys_state_check - obtain css for (task, subsys) w/ extra access conds
683 * @task: the target task
684 * @subsys_id: the target subsystem ID
685 * @__c: extra condition expression to be passed to rcu_dereference_check()
687 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
688 * synchronization rules are the same as task_css_set_check().
690 #define task_subsys_state_check(task, subsys_id, __c) \
691 task_css_set_check((task), (__c))->subsys[(subsys_id)]
694 * task_css_set - obtain a task's css_set
695 * @task: the task to obtain css_set for
697 * See task_css_set_check().
699 static inline struct css_set *task_css_set(struct task_struct *task)
701 return task_css_set_check(task, false);
705 * task_subsys_state - obtain css for (task, subsys)
706 * @task: the target task
707 * @subsys_id: the target subsystem ID
709 * See task_subsys_state_check().
711 static inline struct cgroup_subsys_state *
712 task_subsys_state(struct task_struct *task, int subsys_id)
714 return task_subsys_state_check(task, subsys_id, false);
717 static inline struct cgroup* task_cgroup(struct task_struct *task,
720 return task_subsys_state(task, subsys_id)->cgroup;
723 struct cgroup *cgroup_next_sibling(struct cgroup *pos);
726 * cgroup_for_each_child - iterate through children of a cgroup
727 * @pos: the cgroup * to use as the loop cursor
728 * @cgrp: cgroup whose children to walk
730 * Walk @cgrp's children. Must be called under rcu_read_lock(). A child
731 * cgroup which hasn't finished ->css_online() or already has finished
732 * ->css_offline() may show up during traversal and it's each subsystem's
733 * responsibility to verify that each @pos is alive.
735 * If a subsystem synchronizes against the parent in its ->css_online() and
736 * before starting iterating, a cgroup which finished ->css_online() is
737 * guaranteed to be visible in the future iterations.
739 * It is allowed to temporarily drop RCU read lock during iteration. The
740 * caller is responsible for ensuring that @pos remains accessible until
741 * the start of the next iteration by, for example, bumping the css refcnt.
743 #define cgroup_for_each_child(pos, cgrp) \
744 for ((pos) = list_first_or_null_rcu(&(cgrp)->children, \
745 struct cgroup, sibling); \
746 (pos); (pos) = cgroup_next_sibling((pos)))
748 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
749 struct cgroup *cgroup);
750 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
753 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
754 * @pos: the cgroup * to use as the loop cursor
755 * @cgroup: cgroup whose descendants to walk
757 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
758 * descendant cgroup which hasn't finished ->css_online() or already has
759 * finished ->css_offline() may show up during traversal and it's each
760 * subsystem's responsibility to verify that each @pos is alive.
762 * If a subsystem synchronizes against the parent in its ->css_online() and
763 * before starting iterating, and synchronizes against @pos on each
764 * iteration, any descendant cgroup which finished ->css_online() is
765 * guaranteed to be visible in the future iterations.
767 * In other words, the following guarantees that a descendant can't escape
768 * state updates of its ancestors.
772 * Lock @cgrp->parent and @cgrp;
773 * Inherit state from @cgrp->parent;
777 * my_update_state(@cgrp)
780 * Update @cgrp's state;
783 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
785 * Verify @pos is alive and inherit state from @pos->parent;
790 * As long as the inheriting step, including checking the parent state, is
791 * enclosed inside @pos locking, double-locking the parent isn't necessary
792 * while inheriting. The state update to the parent is guaranteed to be
793 * visible by walking order and, as long as inheriting operations to the
794 * same @pos are atomic to each other, multiple updates racing each other
795 * still result in the correct state. It's guaranateed that at least one
796 * inheritance happens for any cgroup after the latest update to its
799 * If checking parent's state requires locking the parent, each inheriting
800 * iteration should lock and unlock both @pos->parent and @pos.
802 * Alternatively, a subsystem may choose to use a single global lock to
803 * synchronize ->css_online() and ->css_offline() against tree-walking
806 * It is allowed to temporarily drop RCU read lock during iteration. The
807 * caller is responsible for ensuring that @pos remains accessible until
808 * the start of the next iteration by, for example, bumping the css refcnt.
810 #define cgroup_for_each_descendant_pre(pos, cgroup) \
811 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
812 pos = cgroup_next_descendant_pre((pos), (cgroup)))
814 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
815 struct cgroup *cgroup);
818 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
819 * @pos: the cgroup * to use as the loop cursor
820 * @cgroup: cgroup whose descendants to walk
822 * Similar to cgroup_for_each_descendant_pre() but performs post-order
823 * traversal instead. Note that the walk visibility guarantee described in
824 * pre-order walk doesn't apply the same to post-order walks.
826 #define cgroup_for_each_descendant_post(pos, cgroup) \
827 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
828 pos = cgroup_next_descendant_post((pos), (cgroup)))
830 /* A cgroup_iter should be treated as an opaque object */
832 struct list_head *cset_link;
833 struct list_head *task;
837 * To iterate across the tasks in a cgroup:
839 * 1) call cgroup_iter_start to initialize an iterator
841 * 2) call cgroup_iter_next() to retrieve member tasks until it
842 * returns NULL or until you want to end the iteration
844 * 3) call cgroup_iter_end() to destroy the iterator.
846 * Or, call cgroup_scan_tasks() to iterate through every task in a
847 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
848 * the test_task() callback, but not while calling the process_task()
851 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
852 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
853 struct cgroup_iter *it);
854 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
855 int cgroup_scan_tasks(struct cgroup_scanner *scan);
856 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
857 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
860 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
861 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
862 * CSS ID is assigned at cgroup allocation (create) automatically
863 * and removed when subsys calls free_css_id() function. This is because
864 * the lifetime of cgroup_subsys_state is subsys's matter.
866 * Looking up and scanning function should be called under rcu_read_lock().
867 * Taking cgroup_mutex is not necessary for following calls.
868 * But the css returned by this routine can be "not populated yet" or "being
869 * destroyed". The caller should check css and cgroup's status.
873 * Typically Called at ->destroy(), or somewhere the subsys frees
874 * cgroup_subsys_state.
876 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
878 /* Find a cgroup_subsys_state which has given ID */
880 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
882 /* Returns true if root is ancestor of cg */
883 bool css_is_ancestor(struct cgroup_subsys_state *cg,
884 const struct cgroup_subsys_state *root);
886 /* Get id and depth of css */
887 unsigned short css_id(struct cgroup_subsys_state *css);
888 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
890 #else /* !CONFIG_CGROUPS */
892 static inline int cgroup_init_early(void) { return 0; }
893 static inline int cgroup_init(void) { return 0; }
894 static inline void cgroup_fork(struct task_struct *p) {}
895 static inline void cgroup_post_fork(struct task_struct *p) {}
896 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
898 static inline int cgroupstats_build(struct cgroupstats *stats,
899 struct dentry *dentry)
904 /* No cgroups - nothing to do */
905 static inline int cgroup_attach_task_all(struct task_struct *from,
906 struct task_struct *t)
911 #endif /* !CONFIG_CGROUPS */
913 #endif /* _LINUX_CGROUP_H */