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>
19 #include <linux/workqueue.h>
20 #include <linux/xattr.h>
30 extern int cgroup_init_early(void);
31 extern int cgroup_init(void);
32 extern void cgroup_lock(void);
33 extern int cgroup_lock_is_held(void);
34 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
35 extern void cgroup_unlock(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 const struct file_operations proc_cgroup_operations;
46 /* Define the enumeration of all builtin cgroup subsystems */
47 #define SUBSYS(_x) _x ## _subsys_id,
48 #define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
49 enum cgroup_subsys_id {
50 #include <linux/cgroup_subsys.h>
53 #undef IS_SUBSYS_ENABLED
56 /* Per-subsystem/per-cgroup state maintained by the system. */
57 struct cgroup_subsys_state {
59 * The cgroup that this subsystem is attached to. Useful
60 * for subsystems that want to know about the cgroup
63 struct cgroup *cgroup;
66 * State maintained by the cgroup system to allow subsystems
67 * to be "busy". Should be accessed via css_get(),
68 * css_tryget() and and css_put().
74 /* ID for this css, if possible */
75 struct css_id __rcu *id;
77 /* Used to put @cgroup->dentry on the last css_put() */
78 struct work_struct dput_work;
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 */
85 CSS_CLEAR_CSS_REFS, /* @ss->__DEPRECATED_clear_css_refs */
88 /* Caller must verify that the css is not for root cgroup */
89 static inline void __css_get(struct cgroup_subsys_state *css, int count)
91 atomic_add(count, &css->refcnt);
95 * Call css_get() to hold a reference on the css; it can be used
96 * for a reference obtained via:
97 * - an existing ref-counted reference to the css
98 * - task->cgroups for a locked task
101 static inline void css_get(struct cgroup_subsys_state *css)
103 /* We don't need to reference count the root state */
104 if (!test_bit(CSS_ROOT, &css->flags))
108 static inline bool css_is_removed(struct cgroup_subsys_state *css)
110 return test_bit(CSS_REMOVED, &css->flags);
114 * Call css_tryget() to take a reference on a css if your existing
115 * (known-valid) reference isn't already ref-counted. Returns false if
116 * the css has been destroyed.
119 extern bool __css_tryget(struct cgroup_subsys_state *css);
120 static inline bool css_tryget(struct cgroup_subsys_state *css)
122 if (test_bit(CSS_ROOT, &css->flags))
124 return __css_tryget(css);
128 * css_put() should be called to release a reference taken by
129 * css_get() or css_tryget()
132 extern void __css_put(struct cgroup_subsys_state *css);
133 static inline void css_put(struct cgroup_subsys_state *css)
135 if (!test_bit(CSS_ROOT, &css->flags))
139 /* bits in struct cgroup flags field */
141 /* Control Group is dead */
144 * Control Group has previously had a child cgroup or a task,
145 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
148 /* Control Group requires release notifications to userspace */
149 CGRP_NOTIFY_ON_RELEASE,
151 * A thread in rmdir() is wating for this cgroup.
155 * Clone cgroup values when creating a new child cgroup
161 unsigned long flags; /* "unsigned long" so bitops work */
164 * count users of this cgroup. >0 means busy, but doesn't
165 * necessarily indicate the number of tasks in the cgroup
170 * We link our 'sibling' struct into our parent's 'children'.
171 * Our children link their 'sibling' into our 'children'.
173 struct list_head sibling; /* my parent's children */
174 struct list_head children; /* my children */
175 struct list_head files; /* my files */
177 struct cgroup *parent; /* my parent */
178 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
180 /* Private pointers for each registered subsystem */
181 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
183 struct cgroupfs_root *root;
184 struct cgroup *top_cgroup;
187 * List of cg_cgroup_links pointing at css_sets with
188 * tasks in this cgroup. Protected by css_set_lock
190 struct list_head css_sets;
192 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
193 struct list_head cft_q_node; /* used during cftype add/rm */
196 * Linked list running through all cgroups that can
197 * potentially be reaped by the release agent. Protected by
200 struct list_head release_list;
203 * list of pidlists, up to two for each namespace (one for procs, one
204 * for tasks); created on demand.
206 struct list_head pidlists;
207 struct mutex pidlist_mutex;
209 /* For RCU-protected deletion */
210 struct rcu_head rcu_head;
212 /* List of events which userspace want to receive */
213 struct list_head event_list;
214 spinlock_t event_list_lock;
216 /* directory xattrs */
217 struct simple_xattrs xattrs;
221 * A css_set is a structure holding pointers to a set of
222 * cgroup_subsys_state objects. This saves space in the task struct
223 * object and speeds up fork()/exit(), since a single inc/dec and a
224 * list_add()/del() can bump the reference count on the entire cgroup
230 /* Reference count */
234 * List running through all cgroup groups in the same hash
235 * slot. Protected by css_set_lock
237 struct hlist_node hlist;
240 * List running through all tasks using this cgroup
241 * group. Protected by css_set_lock
243 struct list_head tasks;
246 * List of cg_cgroup_link objects on link chains from
247 * cgroups referenced from this css_set. Protected by
250 struct list_head cg_links;
253 * Set of subsystem states, one for each subsystem. This array
254 * is immutable after creation apart from the init_css_set
255 * during subsystem registration (at boot time) and modular subsystem
258 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
260 /* For RCU-protected deletion */
261 struct rcu_head rcu_head;
265 * cgroup_map_cb is an abstract callback API for reporting map-valued
269 struct cgroup_map_cb {
270 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
275 * struct cftype: handler definitions for cgroup control files
277 * When reading/writing to a file:
278 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
279 * - the 'cftype' of the file is file->f_dentry->d_fsdata
283 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
284 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
286 #define MAX_CFTYPE_NAME 64
290 * By convention, the name should begin with the name of the
291 * subsystem, followed by a period. Zero length string indicates
292 * end of cftype array.
294 char name[MAX_CFTYPE_NAME];
297 * If not 0, file mode is set to this value, otherwise it will
298 * be figured out automatically
303 * If non-zero, defines the maximum length of string that can
304 * be passed to write_string; defaults to 64
306 size_t max_write_len;
312 struct simple_xattrs xattrs;
314 int (*open)(struct inode *inode, struct file *file);
315 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
317 char __user *buf, size_t nbytes, loff_t *ppos);
319 * read_u64() is a shortcut for the common case of returning a
320 * single integer. Use it in place of read()
322 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
324 * read_s64() is a signed version of read_u64()
326 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
328 * read_map() is used for defining a map of key/value
329 * pairs. It should call cb->fill(cb, key, value) for each
330 * entry. The key/value pairs (and their ordering) should not
331 * change between reboots.
333 int (*read_map)(struct cgroup *cont, struct cftype *cft,
334 struct cgroup_map_cb *cb);
336 * read_seq_string() is used for outputting a simple sequence
339 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
342 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
344 const char __user *buf, size_t nbytes, loff_t *ppos);
347 * write_u64() is a shortcut for the common case of accepting
348 * a single integer (as parsed by simple_strtoull) from
349 * userspace. Use in place of write(); return 0 or error.
351 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
353 * write_s64() is a signed version of write_u64()
355 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
358 * write_string() is passed a nul-terminated kernelspace
359 * buffer of maximum length determined by max_write_len.
360 * Returns 0 or -ve error code.
362 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
365 * trigger() callback can be used to get some kick from the
366 * userspace, when the actual string written is not important
367 * at all. The private field can be used to determine the
368 * kick type for multiplexing.
370 int (*trigger)(struct cgroup *cgrp, unsigned int event);
372 int (*release)(struct inode *inode, struct file *file);
375 * register_event() callback will be used to add new userspace
376 * waiter for changes related to the cftype. Implement it if
377 * you want to provide this functionality. Use eventfd_signal()
378 * on eventfd to send notification to userspace.
380 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
381 struct eventfd_ctx *eventfd, const char *args);
383 * unregister_event() callback will be called when userspace
384 * closes the eventfd or on cgroup removing.
385 * This callback must be implemented, if you want provide
386 * notification functionality.
388 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
389 struct eventfd_ctx *eventfd);
393 * cftype_sets describe cftypes belonging to a subsystem and are chained at
394 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
395 * terminated by zero length name.
398 struct list_head node; /* chained at subsys->cftsets */
402 struct cgroup_scanner {
404 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
405 void (*process_task)(struct task_struct *p,
406 struct cgroup_scanner *scan);
407 struct ptr_heap *heap;
411 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
412 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
414 int cgroup_is_removed(const struct cgroup *cgrp);
416 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
418 int cgroup_task_count(const struct cgroup *cgrp);
420 /* Return true if cgrp is a descendant of the task's cgroup */
421 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
424 * When the subsys has to access css and may add permanent refcnt to css,
425 * it should take care of racy conditions with rmdir(). Following set of
426 * functions, is for stop/restart rmdir if necessary.
427 * Because these will call css_get/put, "css" should be alive css.
429 * cgroup_exclude_rmdir();
430 * ...do some jobs which may access arbitrary empty cgroup
431 * cgroup_release_and_wakeup_rmdir();
433 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
434 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
437 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
438 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
441 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
444 struct cgroup_taskset;
445 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
446 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
447 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
448 int cgroup_taskset_size(struct cgroup_taskset *tset);
451 * cgroup_taskset_for_each - iterate cgroup_taskset
452 * @task: the loop cursor
453 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
454 * @tset: taskset to iterate
456 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
457 for ((task) = cgroup_taskset_first((tset)); (task); \
458 (task) = cgroup_taskset_next((tset))) \
459 if (!(skip_cgrp) || \
460 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
463 * Control Group subsystem type.
464 * See Documentation/cgroups/cgroups.txt for details
467 struct cgroup_subsys {
468 struct cgroup_subsys_state *(*create)(struct cgroup *cgrp);
469 int (*pre_destroy)(struct cgroup *cgrp);
470 void (*destroy)(struct cgroup *cgrp);
471 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
472 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
473 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
474 void (*fork)(struct task_struct *task);
475 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
476 struct task_struct *task);
477 void (*post_clone)(struct cgroup *cgrp);
478 void (*bind)(struct cgroup *root);
485 * True if this subsys uses ID. ID is not available before cgroup_init()
486 * (not available in early_init time.)
491 * If %true, cgroup removal will try to clear css refs by retrying
492 * ss->pre_destroy() until there's no css ref left. This behavior
493 * is strictly for backward compatibility and will be removed as
494 * soon as the current user (memcg) is updated.
496 * If %false, ss->pre_destroy() can't fail and cgroup removal won't
497 * wait for css refs to drop to zero before proceeding.
499 bool __DEPRECATED_clear_css_refs;
502 * If %false, this subsystem is properly hierarchical -
503 * configuration, resource accounting and restriction on a parent
504 * cgroup cover those of its children. If %true, hierarchy support
505 * is broken in some ways - some subsystems ignore hierarchy
506 * completely while others are only implemented half-way.
508 * It's now disallowed to create nested cgroups if the subsystem is
509 * broken and cgroup core will emit a warning message on such
510 * cases. Eventually, all subsystems will be made properly
511 * hierarchical and this will go away.
513 bool broken_hierarchy;
514 bool warned_broken_hierarchy;
516 #define MAX_CGROUP_TYPE_NAMELEN 32
520 * Link to parent, and list entry in parent's children.
521 * Protected by cgroup_lock()
523 struct cgroupfs_root *root;
524 struct list_head sibling;
525 /* used when use_id == true */
529 /* list of cftype_sets */
530 struct list_head cftsets;
532 /* base cftypes, automatically [de]registered with subsys itself */
533 struct cftype *base_cftypes;
534 struct cftype_set base_cftset;
536 /* should be defined only by modular subsystems */
537 struct module *module;
540 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
541 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
542 #include <linux/cgroup_subsys.h>
543 #undef IS_SUBSYS_ENABLED
546 static inline struct cgroup_subsys_state *cgroup_subsys_state(
547 struct cgroup *cgrp, int subsys_id)
549 return cgrp->subsys[subsys_id];
553 * function to get the cgroup_subsys_state which allows for extra
554 * rcu_dereference_check() conditions, such as locks used during the
555 * cgroup_subsys::attach() methods.
557 #define task_subsys_state_check(task, subsys_id, __c) \
558 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
559 lockdep_is_held(&task->alloc_lock) || \
560 cgroup_lock_is_held() || (__c))
562 static inline struct cgroup_subsys_state *
563 task_subsys_state(struct task_struct *task, int subsys_id)
565 return task_subsys_state_check(task, subsys_id, false);
568 static inline struct cgroup* task_cgroup(struct task_struct *task,
571 return task_subsys_state(task, subsys_id)->cgroup;
574 /* A cgroup_iter should be treated as an opaque object */
576 struct list_head *cg_link;
577 struct list_head *task;
581 * To iterate across the tasks in a cgroup:
583 * 1) call cgroup_iter_start to initialize an iterator
585 * 2) call cgroup_iter_next() to retrieve member tasks until it
586 * returns NULL or until you want to end the iteration
588 * 3) call cgroup_iter_end() to destroy the iterator.
590 * Or, call cgroup_scan_tasks() to iterate through every task in a
591 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
592 * the test_task() callback, but not while calling the process_task()
595 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
596 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
597 struct cgroup_iter *it);
598 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
599 int cgroup_scan_tasks(struct cgroup_scanner *scan);
600 int cgroup_attach_task(struct cgroup *, struct task_struct *);
601 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
604 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
605 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
606 * CSS ID is assigned at cgroup allocation (create) automatically
607 * and removed when subsys calls free_css_id() function. This is because
608 * the lifetime of cgroup_subsys_state is subsys's matter.
610 * Looking up and scanning function should be called under rcu_read_lock().
611 * Taking cgroup_mutex is not necessary for following calls.
612 * But the css returned by this routine can be "not populated yet" or "being
613 * destroyed". The caller should check css and cgroup's status.
617 * Typically Called at ->destroy(), or somewhere the subsys frees
618 * cgroup_subsys_state.
620 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
622 /* Find a cgroup_subsys_state which has given ID */
624 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
627 * Get a cgroup whose id is greater than or equal to id under tree of root.
628 * Returning a cgroup_subsys_state or NULL.
630 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
631 struct cgroup_subsys_state *root, int *foundid);
633 /* Returns true if root is ancestor of cg */
634 bool css_is_ancestor(struct cgroup_subsys_state *cg,
635 const struct cgroup_subsys_state *root);
637 /* Get id and depth of css */
638 unsigned short css_id(struct cgroup_subsys_state *css);
639 unsigned short css_depth(struct cgroup_subsys_state *css);
640 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
642 #else /* !CONFIG_CGROUPS */
644 static inline int cgroup_init_early(void) { return 0; }
645 static inline int cgroup_init(void) { return 0; }
646 static inline void cgroup_fork(struct task_struct *p) {}
647 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
648 static inline void cgroup_post_fork(struct task_struct *p) {}
649 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
651 static inline void cgroup_lock(void) {}
652 static inline void cgroup_unlock(void) {}
653 static inline int cgroupstats_build(struct cgroupstats *stats,
654 struct dentry *dentry)
659 /* No cgroups - nothing to do */
660 static inline int cgroup_attach_task_all(struct task_struct *from,
661 struct task_struct *t)
666 #endif /* !CONFIG_CGROUPS */
668 #endif /* _LINUX_CGROUP_H */