Merge branch 'for-3.11/core' of git://git.kernel.dk/linux-block
[linux-2.6.git] / include / linux / cgroup.h
blob297462b9f41acb7c95607daa36516b017e37be53
1 #ifndef _LINUX_CGROUP_H
2 #define _LINUX_CGROUP_H
3 /*
4 * cgroup interface
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
9 */
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>
22 #include <linux/fs.h>
23 #include <linux/percpu-refcount.h>
25 #ifdef CONFIG_CGROUPS
27 struct cgroupfs_root;
28 struct cgroup_subsys;
29 struct inode;
30 struct cgroup;
31 struct css_id;
32 struct eventfd_ctx;
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
63 CGROUP_SUBSYS_COUNT,
65 #undef SUBSYS
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
72 * hierarchy structure
74 struct cgroup *cgroup;
76 /* reference count - access via css_[try]get() and css_put() */
77 struct percpu_ref refcnt;
79 unsigned long flags;
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 */
88 enum {
89 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
90 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
93 /**
94 * css_get - obtain a reference on the specified css
95 * @css: target 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
108 * @css: target 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;
114 * %false otherwise.
116 static inline bool css_tryget(struct cgroup_subsys_state *css)
118 if (css->flags & CSS_ROOT)
119 return true;
120 return percpu_ref_tryget(&css->refcnt);
124 * css_put - put a css reference
125 * @css: target css
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 */
136 enum {
137 /* Control Group is dead */
138 CGRP_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)
143 CGRP_RELEASABLE,
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 */
153 CGRP_SANE_BEHAVIOR,
156 struct cgroup_name {
157 struct rcu_head rcu_head;
158 char name[];
161 struct cgroup {
162 unsigned long flags; /* "unsigned long" so bitops work */
164 int id; /* ida allocated in-hierarchy ID */
167 * We link our 'sibling' struct into our parent's 'children'.
168 * Our children link their 'sibling' into our 'children'.
170 struct list_head sibling; /* my parent's children */
171 struct list_head children; /* my children */
172 struct list_head files; /* my files */
174 struct cgroup *parent; /* my parent */
175 struct dentry *dentry; /* cgroup fs entry, RCU protected */
178 * Monotonically increasing unique serial number which defines a
179 * uniform order among all cgroups. It's guaranteed that all
180 * ->children lists are in the ascending order of ->serial_nr.
181 * It's used to allow interrupting and resuming iterations.
183 u64 serial_nr;
186 * This is a copy of dentry->d_name, and it's needed because
187 * we can't use dentry->d_name in cgroup_path().
189 * You must acquire rcu_read_lock() to access cgrp->name, and
190 * the only place that can change it is rename(), which is
191 * protected by parent dir's i_mutex.
193 * Normally you should use cgroup_name() wrapper rather than
194 * access it directly.
196 struct cgroup_name __rcu *name;
198 /* Private pointers for each registered subsystem */
199 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
201 struct cgroupfs_root *root;
204 * List of cgrp_cset_links pointing at css_sets with tasks in this
205 * cgroup. Protected by css_set_lock.
207 struct list_head cset_links;
210 * Linked list running through all cgroups that can
211 * potentially be reaped by the release agent. Protected by
212 * release_list_lock
214 struct list_head release_list;
217 * list of pidlists, up to two for each namespace (one for procs, one
218 * for tasks); created on demand.
220 struct list_head pidlists;
221 struct mutex pidlist_mutex;
223 /* For css percpu_ref killing and RCU-protected deletion */
224 struct rcu_head rcu_head;
225 struct work_struct destroy_work;
226 atomic_t css_kill_cnt;
228 /* List of events which userspace want to receive */
229 struct list_head event_list;
230 spinlock_t event_list_lock;
232 /* directory xattrs */
233 struct simple_xattrs xattrs;
236 #define MAX_CGROUP_ROOT_NAMELEN 64
238 /* cgroupfs_root->flags */
239 enum {
241 * Unfortunately, cgroup core and various controllers are riddled
242 * with idiosyncrasies and pointless options. The following flag,
243 * when set, will force sane behavior - some options are forced on,
244 * others are disallowed, and some controllers will change their
245 * hierarchical or other behaviors.
247 * The set of behaviors affected by this flag are still being
248 * determined and developed and the mount option for this flag is
249 * prefixed with __DEVEL__. The prefix will be dropped once we
250 * reach the point where all behaviors are compatible with the
251 * planned unified hierarchy, which will automatically turn on this
252 * flag.
254 * The followings are the behaviors currently affected this flag.
256 * - Mount options "noprefix" and "clone_children" are disallowed.
257 * Also, cgroupfs file cgroup.clone_children is not created.
259 * - When mounting an existing superblock, mount options should
260 * match.
262 * - Remount is disallowed.
264 * - rename(2) is disallowed.
266 * - "tasks" is removed. Everything should be at process
267 * granularity. Use "cgroup.procs" instead.
269 * - "release_agent" and "notify_on_release" are removed.
270 * Replacement notification mechanism will be implemented.
272 * - cpuset: tasks will be kept in empty cpusets when hotplug happens
273 * and take masks of ancestors with non-empty cpus/mems, instead of
274 * being moved to an ancestor.
276 * - cpuset: a task can be moved into an empty cpuset, and again it
277 * takes masks of ancestors.
279 * - memcg: use_hierarchy is on by default and the cgroup file for
280 * the flag is not created.
282 * - blkcg: blk-throttle becomes properly hierarchical.
284 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0),
286 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
287 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
289 /* mount options live below bit 16 */
290 CGRP_ROOT_OPTION_MASK = (1 << 16) - 1,
292 CGRP_ROOT_SUBSYS_BOUND = (1 << 16), /* subsystems finished binding */
296 * A cgroupfs_root represents the root of a cgroup hierarchy, and may be
297 * associated with a superblock to form an active hierarchy. This is
298 * internal to cgroup core. Don't access directly from controllers.
300 struct cgroupfs_root {
301 struct super_block *sb;
303 /* The bitmask of subsystems attached to this hierarchy */
304 unsigned long subsys_mask;
306 /* Unique id for this hierarchy. */
307 int hierarchy_id;
309 /* A list running through the attached subsystems */
310 struct list_head subsys_list;
312 /* The root cgroup for this hierarchy */
313 struct cgroup top_cgroup;
315 /* Tracks how many cgroups are currently defined in hierarchy.*/
316 int number_of_cgroups;
318 /* A list running through the active hierarchies */
319 struct list_head root_list;
321 /* Hierarchy-specific flags */
322 unsigned long flags;
324 /* IDs for cgroups in this hierarchy */
325 struct ida cgroup_ida;
327 /* The path to use for release notifications. */
328 char release_agent_path[PATH_MAX];
330 /* The name for this hierarchy - may be empty */
331 char name[MAX_CGROUP_ROOT_NAMELEN];
335 * A css_set is a structure holding pointers to a set of
336 * cgroup_subsys_state objects. This saves space in the task struct
337 * object and speeds up fork()/exit(), since a single inc/dec and a
338 * list_add()/del() can bump the reference count on the entire cgroup
339 * set for a task.
342 struct css_set {
344 /* Reference count */
345 atomic_t refcount;
348 * List running through all cgroup groups in the same hash
349 * slot. Protected by css_set_lock
351 struct hlist_node hlist;
354 * List running through all tasks using this cgroup
355 * group. Protected by css_set_lock
357 struct list_head tasks;
360 * List of cgrp_cset_links pointing at cgroups referenced from this
361 * css_set. Protected by css_set_lock.
363 struct list_head cgrp_links;
366 * Set of subsystem states, one for each subsystem. This array
367 * is immutable after creation apart from the init_css_set
368 * during subsystem registration (at boot time) and modular subsystem
369 * loading/unloading.
371 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
373 /* For RCU-protected deletion */
374 struct rcu_head rcu_head;
378 * cgroup_map_cb is an abstract callback API for reporting map-valued
379 * control files
382 struct cgroup_map_cb {
383 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
384 void *state;
388 * struct cftype: handler definitions for cgroup control files
390 * When reading/writing to a file:
391 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
392 * - the 'cftype' of the file is file->f_dentry->d_fsdata
395 /* cftype->flags */
396 enum {
397 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cg */
398 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cg */
399 CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */
402 #define MAX_CFTYPE_NAME 64
404 struct cftype {
406 * By convention, the name should begin with the name of the
407 * subsystem, followed by a period. Zero length string indicates
408 * end of cftype array.
410 char name[MAX_CFTYPE_NAME];
411 int private;
413 * If not 0, file mode is set to this value, otherwise it will
414 * be figured out automatically
416 umode_t mode;
419 * If non-zero, defines the maximum length of string that can
420 * be passed to write_string; defaults to 64
422 size_t max_write_len;
424 /* CFTYPE_* flags */
425 unsigned int flags;
427 int (*open)(struct inode *inode, struct file *file);
428 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
429 struct file *file,
430 char __user *buf, size_t nbytes, loff_t *ppos);
432 * read_u64() is a shortcut for the common case of returning a
433 * single integer. Use it in place of read()
435 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
437 * read_s64() is a signed version of read_u64()
439 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
441 * read_map() is used for defining a map of key/value
442 * pairs. It should call cb->fill(cb, key, value) for each
443 * entry. The key/value pairs (and their ordering) should not
444 * change between reboots.
446 int (*read_map)(struct cgroup *cgrp, struct cftype *cft,
447 struct cgroup_map_cb *cb);
449 * read_seq_string() is used for outputting a simple sequence
450 * using seqfile.
452 int (*read_seq_string)(struct cgroup *cgrp, struct cftype *cft,
453 struct seq_file *m);
455 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
456 struct file *file,
457 const char __user *buf, size_t nbytes, loff_t *ppos);
460 * write_u64() is a shortcut for the common case of accepting
461 * a single integer (as parsed by simple_strtoull) from
462 * userspace. Use in place of write(); return 0 or error.
464 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
466 * write_s64() is a signed version of write_u64()
468 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
471 * write_string() is passed a nul-terminated kernelspace
472 * buffer of maximum length determined by max_write_len.
473 * Returns 0 or -ve error code.
475 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
476 const char *buffer);
478 * trigger() callback can be used to get some kick from the
479 * userspace, when the actual string written is not important
480 * at all. The private field can be used to determine the
481 * kick type for multiplexing.
483 int (*trigger)(struct cgroup *cgrp, unsigned int event);
485 int (*release)(struct inode *inode, struct file *file);
488 * register_event() callback will be used to add new userspace
489 * waiter for changes related to the cftype. Implement it if
490 * you want to provide this functionality. Use eventfd_signal()
491 * on eventfd to send notification to userspace.
493 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
494 struct eventfd_ctx *eventfd, const char *args);
496 * unregister_event() callback will be called when userspace
497 * closes the eventfd or on cgroup removing.
498 * This callback must be implemented, if you want provide
499 * notification functionality.
501 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
502 struct eventfd_ctx *eventfd);
506 * cftype_sets describe cftypes belonging to a subsystem and are chained at
507 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
508 * terminated by zero length name.
510 struct cftype_set {
511 struct list_head node; /* chained at subsys->cftsets */
512 struct cftype *cfts;
515 struct cgroup_scanner {
516 struct cgroup *cg;
517 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
518 void (*process_task)(struct task_struct *p,
519 struct cgroup_scanner *scan);
520 struct ptr_heap *heap;
521 void *data;
525 * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This
526 * function can be called as long as @cgrp is accessible.
528 static inline bool cgroup_sane_behavior(const struct cgroup *cgrp)
530 return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR;
533 /* Caller should hold rcu_read_lock() */
534 static inline const char *cgroup_name(const struct cgroup *cgrp)
536 return rcu_dereference(cgrp->name)->name;
539 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
540 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
542 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
544 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
545 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
546 char *buf, size_t buflen);
548 int cgroup_task_count(const struct cgroup *cgrp);
551 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
552 * methods.
554 struct cgroup_taskset;
555 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
556 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
557 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
558 int cgroup_taskset_size(struct cgroup_taskset *tset);
561 * cgroup_taskset_for_each - iterate cgroup_taskset
562 * @task: the loop cursor
563 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
564 * @tset: taskset to iterate
566 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
567 for ((task) = cgroup_taskset_first((tset)); (task); \
568 (task) = cgroup_taskset_next((tset))) \
569 if (!(skip_cgrp) || \
570 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
573 * Control Group subsystem type.
574 * See Documentation/cgroups/cgroups.txt for details
577 struct cgroup_subsys {
578 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
579 int (*css_online)(struct cgroup *cgrp);
580 void (*css_offline)(struct cgroup *cgrp);
581 void (*css_free)(struct cgroup *cgrp);
583 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
584 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
585 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
586 void (*fork)(struct task_struct *task);
587 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
588 struct task_struct *task);
589 void (*bind)(struct cgroup *root);
591 int subsys_id;
592 int disabled;
593 int early_init;
595 * True if this subsys uses ID. ID is not available before cgroup_init()
596 * (not available in early_init time.)
598 bool use_id;
601 * If %false, this subsystem is properly hierarchical -
602 * configuration, resource accounting and restriction on a parent
603 * cgroup cover those of its children. If %true, hierarchy support
604 * is broken in some ways - some subsystems ignore hierarchy
605 * completely while others are only implemented half-way.
607 * It's now disallowed to create nested cgroups if the subsystem is
608 * broken and cgroup core will emit a warning message on such
609 * cases. Eventually, all subsystems will be made properly
610 * hierarchical and this will go away.
612 bool broken_hierarchy;
613 bool warned_broken_hierarchy;
615 #define MAX_CGROUP_TYPE_NAMELEN 32
616 const char *name;
619 * Link to parent, and list entry in parent's children.
620 * Protected by cgroup_lock()
622 struct cgroupfs_root *root;
623 struct list_head sibling;
624 /* used when use_id == true */
625 struct idr idr;
626 spinlock_t id_lock;
628 /* list of cftype_sets */
629 struct list_head cftsets;
631 /* base cftypes, automatically [de]registered with subsys itself */
632 struct cftype *base_cftypes;
633 struct cftype_set base_cftset;
635 /* should be defined only by modular subsystems */
636 struct module *module;
639 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
640 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
641 #include <linux/cgroup_subsys.h>
642 #undef IS_SUBSYS_ENABLED
643 #undef SUBSYS
645 static inline struct cgroup_subsys_state *cgroup_subsys_state(
646 struct cgroup *cgrp, int subsys_id)
648 return cgrp->subsys[subsys_id];
652 * task_css_set_check - obtain a task's css_set with extra access conditions
653 * @task: the task to obtain css_set for
654 * @__c: extra condition expression to be passed to rcu_dereference_check()
656 * A task's css_set is RCU protected, initialized and exited while holding
657 * task_lock(), and can only be modified while holding both cgroup_mutex
658 * and task_lock() while the task is alive. This macro verifies that the
659 * caller is inside proper critical section and returns @task's css_set.
661 * The caller can also specify additional allowed conditions via @__c, such
662 * as locks used during the cgroup_subsys::attach() methods.
664 #ifdef CONFIG_PROVE_RCU
665 extern struct mutex cgroup_mutex;
666 #define task_css_set_check(task, __c) \
667 rcu_dereference_check((task)->cgroups, \
668 lockdep_is_held(&(task)->alloc_lock) || \
669 lockdep_is_held(&cgroup_mutex) || (__c))
670 #else
671 #define task_css_set_check(task, __c) \
672 rcu_dereference((task)->cgroups)
673 #endif
676 * task_subsys_state_check - obtain css for (task, subsys) w/ extra access conds
677 * @task: the target task
678 * @subsys_id: the target subsystem ID
679 * @__c: extra condition expression to be passed to rcu_dereference_check()
681 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
682 * synchronization rules are the same as task_css_set_check().
684 #define task_subsys_state_check(task, subsys_id, __c) \
685 task_css_set_check((task), (__c))->subsys[(subsys_id)]
688 * task_css_set - obtain a task's css_set
689 * @task: the task to obtain css_set for
691 * See task_css_set_check().
693 static inline struct css_set *task_css_set(struct task_struct *task)
695 return task_css_set_check(task, false);
699 * task_subsys_state - obtain css for (task, subsys)
700 * @task: the target task
701 * @subsys_id: the target subsystem ID
703 * See task_subsys_state_check().
705 static inline struct cgroup_subsys_state *
706 task_subsys_state(struct task_struct *task, int subsys_id)
708 return task_subsys_state_check(task, subsys_id, false);
711 static inline struct cgroup* task_cgroup(struct task_struct *task,
712 int subsys_id)
714 return task_subsys_state(task, subsys_id)->cgroup;
717 struct cgroup *cgroup_next_sibling(struct cgroup *pos);
720 * cgroup_for_each_child - iterate through children of a cgroup
721 * @pos: the cgroup * to use as the loop cursor
722 * @cgrp: cgroup whose children to walk
724 * Walk @cgrp's children. Must be called under rcu_read_lock(). A child
725 * cgroup which hasn't finished ->css_online() or already has finished
726 * ->css_offline() may show up during traversal and it's each subsystem's
727 * responsibility to verify that each @pos is alive.
729 * If a subsystem synchronizes against the parent in its ->css_online() and
730 * before starting iterating, a cgroup which finished ->css_online() is
731 * guaranteed to be visible in the future iterations.
733 * It is allowed to temporarily drop RCU read lock during iteration. The
734 * caller is responsible for ensuring that @pos remains accessible until
735 * the start of the next iteration by, for example, bumping the css refcnt.
737 #define cgroup_for_each_child(pos, cgrp) \
738 for ((pos) = list_first_or_null_rcu(&(cgrp)->children, \
739 struct cgroup, sibling); \
740 (pos); (pos) = cgroup_next_sibling((pos)))
742 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
743 struct cgroup *cgroup);
744 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
747 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
748 * @pos: the cgroup * to use as the loop cursor
749 * @cgroup: cgroup whose descendants to walk
751 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
752 * descendant cgroup which hasn't finished ->css_online() or already has
753 * finished ->css_offline() may show up during traversal and it's each
754 * subsystem's responsibility to verify that each @pos is alive.
756 * If a subsystem synchronizes against the parent in its ->css_online() and
757 * before starting iterating, and synchronizes against @pos on each
758 * iteration, any descendant cgroup which finished ->css_online() is
759 * guaranteed to be visible in the future iterations.
761 * In other words, the following guarantees that a descendant can't escape
762 * state updates of its ancestors.
764 * my_online(@cgrp)
766 * Lock @cgrp->parent and @cgrp;
767 * Inherit state from @cgrp->parent;
768 * Unlock both.
771 * my_update_state(@cgrp)
773 * Lock @cgrp;
774 * Update @cgrp's state;
775 * Unlock @cgrp;
777 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
778 * Lock @pos;
779 * Verify @pos is alive and inherit state from @pos->parent;
780 * Unlock @pos;
784 * As long as the inheriting step, including checking the parent state, is
785 * enclosed inside @pos locking, double-locking the parent isn't necessary
786 * while inheriting. The state update to the parent is guaranteed to be
787 * visible by walking order and, as long as inheriting operations to the
788 * same @pos are atomic to each other, multiple updates racing each other
789 * still result in the correct state. It's guaranateed that at least one
790 * inheritance happens for any cgroup after the latest update to its
791 * parent.
793 * If checking parent's state requires locking the parent, each inheriting
794 * iteration should lock and unlock both @pos->parent and @pos.
796 * Alternatively, a subsystem may choose to use a single global lock to
797 * synchronize ->css_online() and ->css_offline() against tree-walking
798 * operations.
800 * It is allowed to temporarily drop RCU read lock during iteration. The
801 * caller is responsible for ensuring that @pos remains accessible until
802 * the start of the next iteration by, for example, bumping the css refcnt.
804 #define cgroup_for_each_descendant_pre(pos, cgroup) \
805 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
806 pos = cgroup_next_descendant_pre((pos), (cgroup)))
808 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
809 struct cgroup *cgroup);
812 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
813 * @pos: the cgroup * to use as the loop cursor
814 * @cgroup: cgroup whose descendants to walk
816 * Similar to cgroup_for_each_descendant_pre() but performs post-order
817 * traversal instead. Note that the walk visibility guarantee described in
818 * pre-order walk doesn't apply the same to post-order walks.
820 #define cgroup_for_each_descendant_post(pos, cgroup) \
821 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
822 pos = cgroup_next_descendant_post((pos), (cgroup)))
824 /* A cgroup_iter should be treated as an opaque object */
825 struct cgroup_iter {
826 struct list_head *cset_link;
827 struct list_head *task;
831 * To iterate across the tasks in a cgroup:
833 * 1) call cgroup_iter_start to initialize an iterator
835 * 2) call cgroup_iter_next() to retrieve member tasks until it
836 * returns NULL or until you want to end the iteration
838 * 3) call cgroup_iter_end() to destroy the iterator.
840 * Or, call cgroup_scan_tasks() to iterate through every task in a
841 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
842 * the test_task() callback, but not while calling the process_task()
843 * callback.
845 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
846 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
847 struct cgroup_iter *it);
848 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
849 int cgroup_scan_tasks(struct cgroup_scanner *scan);
850 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
851 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
854 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
855 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
856 * CSS ID is assigned at cgroup allocation (create) automatically
857 * and removed when subsys calls free_css_id() function. This is because
858 * the lifetime of cgroup_subsys_state is subsys's matter.
860 * Looking up and scanning function should be called under rcu_read_lock().
861 * Taking cgroup_mutex is not necessary for following calls.
862 * But the css returned by this routine can be "not populated yet" or "being
863 * destroyed". The caller should check css and cgroup's status.
867 * Typically Called at ->destroy(), or somewhere the subsys frees
868 * cgroup_subsys_state.
870 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
872 /* Find a cgroup_subsys_state which has given ID */
874 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
876 /* Returns true if root is ancestor of cg */
877 bool css_is_ancestor(struct cgroup_subsys_state *cg,
878 const struct cgroup_subsys_state *root);
880 /* Get id and depth of css */
881 unsigned short css_id(struct cgroup_subsys_state *css);
882 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
884 #else /* !CONFIG_CGROUPS */
886 static inline int cgroup_init_early(void) { return 0; }
887 static inline int cgroup_init(void) { return 0; }
888 static inline void cgroup_fork(struct task_struct *p) {}
889 static inline void cgroup_post_fork(struct task_struct *p) {}
890 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
892 static inline int cgroupstats_build(struct cgroupstats *stats,
893 struct dentry *dentry)
895 return -EINVAL;
898 /* No cgroups - nothing to do */
899 static inline int cgroup_attach_task_all(struct task_struct *from,
900 struct task_struct *t)
902 return 0;
905 #endif /* !CONFIG_CGROUPS */
907 #endif /* _LINUX_CGROUP_H */