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/cgroupstats.h>
16 #include <linux/prio_heap.h>
17 #include <linux/rwsem.h>
18 #include <linux/idr.h>
28 extern int cgroup_init_early(void);
29 extern int cgroup_init(void);
30 extern void cgroup_lock(void);
31 extern int cgroup_lock_is_held(void);
32 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
33 extern void cgroup_unlock(void);
34 extern void cgroup_fork(struct task_struct *p);
35 extern void cgroup_fork_callbacks(struct task_struct *p);
36 extern void cgroup_post_fork(struct task_struct *p);
37 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
38 extern int cgroupstats_build(struct cgroupstats *stats,
39 struct dentry *dentry);
40 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
41 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
43 extern const struct file_operations proc_cgroup_operations;
45 /* Define the enumeration of all builtin cgroup subsystems */
46 #define SUBSYS(_x) _x ## _subsys_id,
47 enum cgroup_subsys_id {
48 #include <linux/cgroup_subsys.h>
49 CGROUP_BUILTIN_SUBSYS_COUNT
53 * This define indicates the maximum number of subsystems that can be loaded
54 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
55 * track of all of them.
57 #define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
59 /* Per-subsystem/per-cgroup state maintained by the system. */
60 struct cgroup_subsys_state {
62 * The cgroup that this subsystem is attached to. Useful
63 * for subsystems that want to know about the cgroup
66 struct cgroup *cgroup;
69 * State maintained by the cgroup system to allow subsystems
70 * to be "busy". Should be accessed via css_get(),
71 * css_tryget() and and css_put().
77 /* ID for this css, if possible */
78 struct css_id __rcu *id;
81 /* bits in struct cgroup_subsys_state flags field */
83 CSS_ROOT, /* This CSS is the root of the subsystem */
84 CSS_REMOVED, /* This CSS is dead */
87 /* Caller must verify that the css is not for root cgroup */
88 static inline void __css_get(struct cgroup_subsys_state *css, int count)
90 atomic_add(count, &css->refcnt);
94 * Call css_get() to hold a reference on the css; it can be used
95 * for a reference obtained via:
96 * - an existing ref-counted reference to the css
97 * - task->cgroups for a locked task
100 static inline void css_get(struct cgroup_subsys_state *css)
102 /* We don't need to reference count the root state */
103 if (!test_bit(CSS_ROOT, &css->flags))
107 static inline bool css_is_removed(struct cgroup_subsys_state *css)
109 return test_bit(CSS_REMOVED, &css->flags);
113 * Call css_tryget() to take a reference on a css if your existing
114 * (known-valid) reference isn't already ref-counted. Returns false if
115 * the css has been destroyed.
118 static inline bool css_tryget(struct cgroup_subsys_state *css)
120 if (test_bit(CSS_ROOT, &css->flags))
122 while (!atomic_inc_not_zero(&css->refcnt)) {
123 if (test_bit(CSS_REMOVED, &css->flags))
131 * css_put() should be called to release a reference taken by
132 * css_get() or css_tryget()
135 extern void __css_put(struct cgroup_subsys_state *css, int count);
136 static inline void css_put(struct cgroup_subsys_state *css)
138 if (!test_bit(CSS_ROOT, &css->flags))
142 /* bits in struct cgroup flags field */
144 /* Control Group is dead */
147 * Control Group has previously had a child cgroup or a task,
148 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
151 /* Control Group requires release notifications to userspace */
152 CGRP_NOTIFY_ON_RELEASE,
154 * A thread in rmdir() is wating for this cgroup.
158 * Clone cgroup values when creating a new child cgroup
163 /* which pidlist file are we talking about? */
164 enum cgroup_filetype {
170 * A pidlist is a list of pids that virtually represents the contents of one
171 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
172 * a pair (one each for procs, tasks) for each pid namespace that's relevant
175 struct cgroup_pidlist {
177 * used to find which pidlist is wanted. doesn't change as long as
178 * this particular list stays in the list.
180 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
183 /* how many elements the above list has */
185 /* how many files are using the current array */
187 /* each of these stored in a list by its cgroup */
188 struct list_head links;
189 /* pointer to the cgroup we belong to, for list removal purposes */
190 struct cgroup *owner;
191 /* protects the other fields */
192 struct rw_semaphore mutex;
196 unsigned long flags; /* "unsigned long" so bitops work */
199 * count users of this cgroup. >0 means busy, but doesn't
200 * necessarily indicate the number of tasks in the cgroup
205 * We link our 'sibling' struct into our parent's 'children'.
206 * Our children link their 'sibling' into our 'children'.
208 struct list_head sibling; /* my parent's children */
209 struct list_head children; /* my children */
211 struct cgroup *parent; /* my parent */
212 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
214 /* Private pointers for each registered subsystem */
215 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
217 struct cgroupfs_root *root;
218 struct cgroup *top_cgroup;
221 * List of cg_cgroup_links pointing at css_sets with
222 * tasks in this cgroup. Protected by css_set_lock
224 struct list_head css_sets;
227 * Linked list running through all cgroups that can
228 * potentially be reaped by the release agent. Protected by
231 struct list_head release_list;
234 * list of pidlists, up to two for each namespace (one for procs, one
235 * for tasks); created on demand.
237 struct list_head pidlists;
238 struct mutex pidlist_mutex;
240 /* For RCU-protected deletion */
241 struct rcu_head rcu_head;
243 /* List of events which userspace want to recieve */
244 struct list_head event_list;
245 spinlock_t event_list_lock;
249 * A css_set is a structure holding pointers to a set of
250 * cgroup_subsys_state objects. This saves space in the task struct
251 * object and speeds up fork()/exit(), since a single inc/dec and a
252 * list_add()/del() can bump the reference count on the entire cgroup
258 /* Reference count */
262 * List running through all cgroup groups in the same hash
263 * slot. Protected by css_set_lock
265 struct hlist_node hlist;
268 * List running through all tasks using this cgroup
269 * group. Protected by css_set_lock
271 struct list_head tasks;
274 * List of cg_cgroup_link objects on link chains from
275 * cgroups referenced from this css_set. Protected by
278 struct list_head cg_links;
281 * Set of subsystem states, one for each subsystem. This array
282 * is immutable after creation apart from the init_css_set
283 * during subsystem registration (at boot time) and modular subsystem
286 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
288 /* For RCU-protected deletion */
289 struct rcu_head rcu_head;
293 * cgroup_map_cb is an abstract callback API for reporting map-valued
297 struct cgroup_map_cb {
298 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
303 * struct cftype: handler definitions for cgroup control files
305 * When reading/writing to a file:
306 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
307 * - the 'cftype' of the file is file->f_dentry->d_fsdata
310 #define MAX_CFTYPE_NAME 64
313 * By convention, the name should begin with the name of the
314 * subsystem, followed by a period
316 char name[MAX_CFTYPE_NAME];
319 * If not 0, file mode is set to this value, otherwise it will
320 * be figured out automatically
325 * If non-zero, defines the maximum length of string that can
326 * be passed to write_string; defaults to 64
328 size_t max_write_len;
330 int (*open)(struct inode *inode, struct file *file);
331 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
333 char __user *buf, size_t nbytes, loff_t *ppos);
335 * read_u64() is a shortcut for the common case of returning a
336 * single integer. Use it in place of read()
338 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
340 * read_s64() is a signed version of read_u64()
342 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
344 * read_map() is used for defining a map of key/value
345 * pairs. It should call cb->fill(cb, key, value) for each
346 * entry. The key/value pairs (and their ordering) should not
347 * change between reboots.
349 int (*read_map)(struct cgroup *cont, struct cftype *cft,
350 struct cgroup_map_cb *cb);
352 * read_seq_string() is used for outputting a simple sequence
355 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
358 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
360 const char __user *buf, size_t nbytes, loff_t *ppos);
363 * write_u64() is a shortcut for the common case of accepting
364 * a single integer (as parsed by simple_strtoull) from
365 * userspace. Use in place of write(); return 0 or error.
367 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
369 * write_s64() is a signed version of write_u64()
371 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
374 * write_string() is passed a nul-terminated kernelspace
375 * buffer of maximum length determined by max_write_len.
376 * Returns 0 or -ve error code.
378 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
381 * trigger() callback can be used to get some kick from the
382 * userspace, when the actual string written is not important
383 * at all. The private field can be used to determine the
384 * kick type for multiplexing.
386 int (*trigger)(struct cgroup *cgrp, unsigned int event);
388 int (*release)(struct inode *inode, struct file *file);
391 * register_event() callback will be used to add new userspace
392 * waiter for changes related to the cftype. Implement it if
393 * you want to provide this functionality. Use eventfd_signal()
394 * on eventfd to send notification to userspace.
396 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
397 struct eventfd_ctx *eventfd, const char *args);
399 * unregister_event() callback will be called when userspace
400 * closes the eventfd or on cgroup removing.
401 * This callback must be implemented, if you want provide
402 * notification functionality.
404 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
405 struct eventfd_ctx *eventfd);
408 struct cgroup_scanner {
410 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
411 void (*process_task)(struct task_struct *p,
412 struct cgroup_scanner *scan);
413 struct ptr_heap *heap;
418 * Add a new file to the given cgroup directory. Should only be
419 * called by subsystems from within a populate() method
421 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
422 const struct cftype *cft);
425 * Add a set of new files to the given cgroup directory. Should
426 * only be called by subsystems from within a populate() method
428 int cgroup_add_files(struct cgroup *cgrp,
429 struct cgroup_subsys *subsys,
430 const struct cftype cft[],
433 int cgroup_is_removed(const struct cgroup *cgrp);
435 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
437 int cgroup_task_count(const struct cgroup *cgrp);
439 /* Return true if cgrp is a descendant of the task's cgroup */
440 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
443 * When the subsys has to access css and may add permanent refcnt to css,
444 * it should take care of racy conditions with rmdir(). Following set of
445 * functions, is for stop/restart rmdir if necessary.
446 * Because these will call css_get/put, "css" should be alive css.
448 * cgroup_exclude_rmdir();
449 * ...do some jobs which may access arbitrary empty cgroup
450 * cgroup_release_and_wakeup_rmdir();
452 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
453 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
456 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
457 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
460 * Control Group subsystem type.
461 * See Documentation/cgroups/cgroups.txt for details
464 struct cgroup_subsys {
465 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
466 struct cgroup *cgrp);
467 int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
468 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
469 int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
470 struct task_struct *tsk, bool threadgroup);
471 void (*cancel_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
472 struct task_struct *tsk, bool threadgroup);
473 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
474 struct cgroup *old_cgrp, struct task_struct *tsk,
476 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
477 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
478 int (*populate)(struct cgroup_subsys *ss,
479 struct cgroup *cgrp);
480 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
481 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
488 * True if this subsys uses ID. ID is not available before cgroup_init()
489 * (not available in early_init time.)
492 #define MAX_CGROUP_TYPE_NAMELEN 32
496 * Protects sibling/children links of cgroups in this
497 * hierarchy, plus protects which hierarchy (or none) the
498 * subsystem is a part of (i.e. root/sibling). To avoid
499 * potential deadlocks, the following operations should not be
500 * undertaken while holding any hierarchy_mutex:
502 * - allocating memory
503 * - initiating hotplug events
505 struct mutex hierarchy_mutex;
506 struct lock_class_key subsys_key;
509 * Link to parent, and list entry in parent's children.
510 * Protected by this->hierarchy_mutex and cgroup_lock()
512 struct cgroupfs_root *root;
513 struct list_head sibling;
514 /* used when use_id == true */
518 /* should be defined only by modular subsystems */
519 struct module *module;
522 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
523 #include <linux/cgroup_subsys.h>
526 static inline struct cgroup_subsys_state *cgroup_subsys_state(
527 struct cgroup *cgrp, int subsys_id)
529 return cgrp->subsys[subsys_id];
533 * function to get the cgroup_subsys_state which allows for extra
534 * rcu_dereference_check() conditions, such as locks used during the
535 * cgroup_subsys::attach() methods.
537 #define task_subsys_state_check(task, subsys_id, __c) \
538 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
539 rcu_read_lock_held() || \
540 lockdep_is_held(&task->alloc_lock) || \
541 cgroup_lock_is_held() || (__c))
543 static inline struct cgroup_subsys_state *
544 task_subsys_state(struct task_struct *task, int subsys_id)
546 return task_subsys_state_check(task, subsys_id, false);
549 static inline struct cgroup* task_cgroup(struct task_struct *task,
552 return task_subsys_state(task, subsys_id)->cgroup;
555 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
558 /* A cgroup_iter should be treated as an opaque object */
560 struct list_head *cg_link;
561 struct list_head *task;
565 * To iterate across the tasks in a cgroup:
567 * 1) call cgroup_iter_start to initialize an iterator
569 * 2) call cgroup_iter_next() to retrieve member tasks until it
570 * returns NULL or until you want to end the iteration
572 * 3) call cgroup_iter_end() to destroy the iterator.
574 * Or, call cgroup_scan_tasks() to iterate through every task in a
575 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
576 * the test_task() callback, but not while calling the process_task()
579 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
580 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
581 struct cgroup_iter *it);
582 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
583 int cgroup_scan_tasks(struct cgroup_scanner *scan);
584 int cgroup_attach_task(struct cgroup *, struct task_struct *);
585 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
587 static inline int cgroup_attach_task_current_cg(struct task_struct *tsk)
589 return cgroup_attach_task_all(current, tsk);
593 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
594 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
595 * CSS ID is assigned at cgroup allocation (create) automatically
596 * and removed when subsys calls free_css_id() function. This is because
597 * the lifetime of cgroup_subsys_state is subsys's matter.
599 * Looking up and scanning function should be called under rcu_read_lock().
600 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
601 * But the css returned by this routine can be "not populated yet" or "being
602 * destroyed". The caller should check css and cgroup's status.
606 * Typically Called at ->destroy(), or somewhere the subsys frees
607 * cgroup_subsys_state.
609 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
611 /* Find a cgroup_subsys_state which has given ID */
613 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
616 * Get a cgroup whose id is greater than or equal to id under tree of root.
617 * Returning a cgroup_subsys_state or NULL.
619 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
620 struct cgroup_subsys_state *root, int *foundid);
622 /* Returns true if root is ancestor of cg */
623 bool css_is_ancestor(struct cgroup_subsys_state *cg,
624 const struct cgroup_subsys_state *root);
626 /* Get id and depth of css */
627 unsigned short css_id(struct cgroup_subsys_state *css);
628 unsigned short css_depth(struct cgroup_subsys_state *css);
630 #else /* !CONFIG_CGROUPS */
632 static inline int cgroup_init_early(void) { return 0; }
633 static inline int cgroup_init(void) { return 0; }
634 static inline void cgroup_fork(struct task_struct *p) {}
635 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
636 static inline void cgroup_post_fork(struct task_struct *p) {}
637 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
639 static inline void cgroup_lock(void) {}
640 static inline void cgroup_unlock(void) {}
641 static inline int cgroupstats_build(struct cgroupstats *stats,
642 struct dentry *dentry)
647 /* No cgroups - nothing to do */
648 static inline int cgroup_attach_task_all(struct task_struct *from,
649 struct task_struct *t)
653 static inline int cgroup_attach_task_current_cg(struct task_struct *t)
658 #endif /* !CONFIG_CGROUPS */
660 #endif /* _LINUX_CGROUP_H */