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_fork_callbacks(struct task_struct *p);
38 extern void cgroup_post_fork(struct task_struct *p);
39 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
40 extern int cgroupstats_build(struct cgroupstats *stats,
41 struct dentry *dentry);
42 extern int cgroup_load_subsys(struct cgroup_subsys *ss);
43 extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
45 extern const struct file_operations proc_cgroup_operations;
47 /* Define the enumeration of all builtin cgroup subsystems */
48 #define SUBSYS(_x) _x ## _subsys_id,
49 enum cgroup_subsys_id {
50 #include <linux/cgroup_subsys.h>
51 __CGROUP_TEMPORARY_PLACEHOLDER
55 * This define indicates the maximum number of subsystems that can be loaded
56 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
57 * track of all of them.
59 #define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
61 /* Per-subsystem/per-cgroup state maintained by the system. */
62 struct cgroup_subsys_state {
64 * The cgroup that this subsystem is attached to. Useful
65 * for subsystems that want to know about the cgroup
68 struct cgroup *cgroup;
71 * State maintained by the cgroup system to allow subsystems
72 * to be "busy". Should be accessed via css_get(),
73 * css_tryget() and and css_put().
79 /* ID for this css, if possible */
80 struct css_id __rcu *id;
82 /* Used to put @cgroup->dentry on the last css_put() */
83 struct work_struct dput_work;
86 /* bits in struct cgroup_subsys_state flags field */
88 CSS_ROOT, /* This CSS is the root of the subsystem */
89 CSS_REMOVED, /* This CSS is dead */
90 CSS_CLEAR_CSS_REFS, /* @ss->__DEPRECATED_clear_css_refs */
93 /* Caller must verify that the css is not for root cgroup */
94 static inline void __css_get(struct cgroup_subsys_state *css, int count)
96 atomic_add(count, &css->refcnt);
100 * Call css_get() to hold a reference on the css; it can be used
101 * for a reference obtained via:
102 * - an existing ref-counted reference to the css
103 * - task->cgroups for a locked task
106 static inline void css_get(struct cgroup_subsys_state *css)
108 /* We don't need to reference count the root state */
109 if (!test_bit(CSS_ROOT, &css->flags))
113 static inline bool css_is_removed(struct cgroup_subsys_state *css)
115 return test_bit(CSS_REMOVED, &css->flags);
119 * Call css_tryget() to take a reference on a css if your existing
120 * (known-valid) reference isn't already ref-counted. Returns false if
121 * the css has been destroyed.
124 extern bool __css_tryget(struct cgroup_subsys_state *css);
125 static inline bool css_tryget(struct cgroup_subsys_state *css)
127 if (test_bit(CSS_ROOT, &css->flags))
129 return __css_tryget(css);
133 * css_put() should be called to release a reference taken by
134 * css_get() or css_tryget()
137 extern void __css_put(struct cgroup_subsys_state *css);
138 static inline void css_put(struct cgroup_subsys_state *css)
140 if (!test_bit(CSS_ROOT, &css->flags))
144 /* bits in struct cgroup flags field */
146 /* Control Group is dead */
149 * Control Group has previously had a child cgroup or a task,
150 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
153 /* Control Group requires release notifications to userspace */
154 CGRP_NOTIFY_ON_RELEASE,
156 * A thread in rmdir() is wating for this cgroup.
160 * Clone cgroup values when creating a new child cgroup
166 unsigned long flags; /* "unsigned long" so bitops work */
169 * count users of this cgroup. >0 means busy, but doesn't
170 * necessarily indicate the number of tasks in the cgroup
175 * We link our 'sibling' struct into our parent's 'children'.
176 * Our children link their 'sibling' into our 'children'.
178 struct list_head sibling; /* my parent's children */
179 struct list_head children; /* my children */
180 struct list_head files; /* my files */
182 struct cgroup *parent; /* my parent */
183 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
185 /* Private pointers for each registered subsystem */
186 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
188 struct cgroupfs_root *root;
189 struct cgroup *top_cgroup;
192 * List of cg_cgroup_links pointing at css_sets with
193 * tasks in this cgroup. Protected by css_set_lock
195 struct list_head css_sets;
197 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
198 struct list_head cft_q_node; /* used during cftype add/rm */
201 * Linked list running through all cgroups that can
202 * potentially be reaped by the release agent. Protected by
205 struct list_head release_list;
208 * list of pidlists, up to two for each namespace (one for procs, one
209 * for tasks); created on demand.
211 struct list_head pidlists;
212 struct mutex pidlist_mutex;
214 /* For RCU-protected deletion */
215 struct rcu_head rcu_head;
217 /* List of events which userspace want to receive */
218 struct list_head event_list;
219 spinlock_t event_list_lock;
221 /* directory xattrs */
222 struct simple_xattrs xattrs;
226 * A css_set is a structure holding pointers to a set of
227 * cgroup_subsys_state objects. This saves space in the task struct
228 * object and speeds up fork()/exit(), since a single inc/dec and a
229 * list_add()/del() can bump the reference count on the entire cgroup
235 /* Reference count */
239 * List running through all cgroup groups in the same hash
240 * slot. Protected by css_set_lock
242 struct hlist_node hlist;
245 * List running through all tasks using this cgroup
246 * group. Protected by css_set_lock
248 struct list_head tasks;
251 * List of cg_cgroup_link objects on link chains from
252 * cgroups referenced from this css_set. Protected by
255 struct list_head cg_links;
258 * Set of subsystem states, one for each subsystem. This array
259 * is immutable after creation apart from the init_css_set
260 * during subsystem registration (at boot time) and modular subsystem
263 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
265 /* For RCU-protected deletion */
266 struct rcu_head rcu_head;
270 * cgroup_map_cb is an abstract callback API for reporting map-valued
274 struct cgroup_map_cb {
275 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
280 * struct cftype: handler definitions for cgroup control files
282 * When reading/writing to a file:
283 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
284 * - the 'cftype' of the file is file->f_dentry->d_fsdata
288 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
289 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
291 #define MAX_CFTYPE_NAME 64
295 * By convention, the name should begin with the name of the
296 * subsystem, followed by a period. Zero length string indicates
297 * end of cftype array.
299 char name[MAX_CFTYPE_NAME];
302 * If not 0, file mode is set to this value, otherwise it will
303 * be figured out automatically
308 * If non-zero, defines the maximum length of string that can
309 * be passed to write_string; defaults to 64
311 size_t max_write_len;
317 struct simple_xattrs xattrs;
319 int (*open)(struct inode *inode, struct file *file);
320 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
322 char __user *buf, size_t nbytes, loff_t *ppos);
324 * read_u64() is a shortcut for the common case of returning a
325 * single integer. Use it in place of read()
327 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
329 * read_s64() is a signed version of read_u64()
331 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
333 * read_map() is used for defining a map of key/value
334 * pairs. It should call cb->fill(cb, key, value) for each
335 * entry. The key/value pairs (and their ordering) should not
336 * change between reboots.
338 int (*read_map)(struct cgroup *cont, struct cftype *cft,
339 struct cgroup_map_cb *cb);
341 * read_seq_string() is used for outputting a simple sequence
344 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
347 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
349 const char __user *buf, size_t nbytes, loff_t *ppos);
352 * write_u64() is a shortcut for the common case of accepting
353 * a single integer (as parsed by simple_strtoull) from
354 * userspace. Use in place of write(); return 0 or error.
356 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
358 * write_s64() is a signed version of write_u64()
360 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
363 * write_string() is passed a nul-terminated kernelspace
364 * buffer of maximum length determined by max_write_len.
365 * Returns 0 or -ve error code.
367 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
370 * trigger() callback can be used to get some kick from the
371 * userspace, when the actual string written is not important
372 * at all. The private field can be used to determine the
373 * kick type for multiplexing.
375 int (*trigger)(struct cgroup *cgrp, unsigned int event);
377 int (*release)(struct inode *inode, struct file *file);
380 * register_event() callback will be used to add new userspace
381 * waiter for changes related to the cftype. Implement it if
382 * you want to provide this functionality. Use eventfd_signal()
383 * on eventfd to send notification to userspace.
385 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
386 struct eventfd_ctx *eventfd, const char *args);
388 * unregister_event() callback will be called when userspace
389 * closes the eventfd or on cgroup removing.
390 * This callback must be implemented, if you want provide
391 * notification functionality.
393 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
394 struct eventfd_ctx *eventfd);
398 * cftype_sets describe cftypes belonging to a subsystem and are chained at
399 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
400 * terminated by zero length name.
403 struct list_head node; /* chained at subsys->cftsets */
407 struct cgroup_scanner {
409 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
410 void (*process_task)(struct task_struct *p,
411 struct cgroup_scanner *scan);
412 struct ptr_heap *heap;
416 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
417 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
419 int cgroup_is_removed(const struct cgroup *cgrp);
421 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
423 int cgroup_task_count(const struct cgroup *cgrp);
425 /* Return true if cgrp is a descendant of the task's cgroup */
426 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
429 * When the subsys has to access css and may add permanent refcnt to css,
430 * it should take care of racy conditions with rmdir(). Following set of
431 * functions, is for stop/restart rmdir if necessary.
432 * Because these will call css_get/put, "css" should be alive css.
434 * cgroup_exclude_rmdir();
435 * ...do some jobs which may access arbitrary empty cgroup
436 * cgroup_release_and_wakeup_rmdir();
438 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
439 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
442 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
443 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
446 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
449 struct cgroup_taskset;
450 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
451 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
452 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
453 int cgroup_taskset_size(struct cgroup_taskset *tset);
456 * cgroup_taskset_for_each - iterate cgroup_taskset
457 * @task: the loop cursor
458 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
459 * @tset: taskset to iterate
461 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
462 for ((task) = cgroup_taskset_first((tset)); (task); \
463 (task) = cgroup_taskset_next((tset))) \
464 if (!(skip_cgrp) || \
465 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
468 * Control Group subsystem type.
469 * See Documentation/cgroups/cgroups.txt for details
472 struct cgroup_subsys {
473 struct cgroup_subsys_state *(*create)(struct cgroup *cgrp);
474 int (*pre_destroy)(struct cgroup *cgrp);
475 void (*destroy)(struct cgroup *cgrp);
476 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
477 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
478 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
479 void (*fork)(struct task_struct *task);
480 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
481 struct task_struct *task);
482 void (*post_clone)(struct cgroup *cgrp);
483 void (*bind)(struct cgroup *root);
490 * True if this subsys uses ID. ID is not available before cgroup_init()
491 * (not available in early_init time.)
496 * If %true, cgroup removal will try to clear css refs by retrying
497 * ss->pre_destroy() until there's no css ref left. This behavior
498 * is strictly for backward compatibility and will be removed as
499 * soon as the current user (memcg) is updated.
501 * If %false, ss->pre_destroy() can't fail and cgroup removal won't
502 * wait for css refs to drop to zero before proceeding.
504 bool __DEPRECATED_clear_css_refs;
506 #define MAX_CGROUP_TYPE_NAMELEN 32
510 * Link to parent, and list entry in parent's children.
511 * Protected by cgroup_lock()
513 struct cgroupfs_root *root;
514 struct list_head sibling;
515 /* used when use_id == true */
519 /* list of cftype_sets */
520 struct list_head cftsets;
522 /* base cftypes, automatically [de]registered with subsys itself */
523 struct cftype *base_cftypes;
524 struct cftype_set base_cftset;
526 /* should be defined only by modular subsystems */
527 struct module *module;
530 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
531 #include <linux/cgroup_subsys.h>
534 static inline struct cgroup_subsys_state *cgroup_subsys_state(
535 struct cgroup *cgrp, int subsys_id)
537 return cgrp->subsys[subsys_id];
541 * function to get the cgroup_subsys_state which allows for extra
542 * rcu_dereference_check() conditions, such as locks used during the
543 * cgroup_subsys::attach() methods.
545 #define task_subsys_state_check(task, subsys_id, __c) \
546 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
547 lockdep_is_held(&task->alloc_lock) || \
548 cgroup_lock_is_held() || (__c))
550 static inline struct cgroup_subsys_state *
551 task_subsys_state(struct task_struct *task, int subsys_id)
553 return task_subsys_state_check(task, subsys_id, false);
556 static inline struct cgroup* task_cgroup(struct task_struct *task,
559 return task_subsys_state(task, subsys_id)->cgroup;
562 /* A cgroup_iter should be treated as an opaque object */
564 struct list_head *cg_link;
565 struct list_head *task;
569 * To iterate across the tasks in a cgroup:
571 * 1) call cgroup_iter_start to initialize an iterator
573 * 2) call cgroup_iter_next() to retrieve member tasks until it
574 * returns NULL or until you want to end the iteration
576 * 3) call cgroup_iter_end() to destroy the iterator.
578 * Or, call cgroup_scan_tasks() to iterate through every task in a
579 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
580 * the test_task() callback, but not while calling the process_task()
583 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
584 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
585 struct cgroup_iter *it);
586 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
587 int cgroup_scan_tasks(struct cgroup_scanner *scan);
588 int cgroup_attach_task(struct cgroup *, struct task_struct *);
589 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
592 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
593 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
594 * CSS ID is assigned at cgroup allocation (create) automatically
595 * and removed when subsys calls free_css_id() function. This is because
596 * the lifetime of cgroup_subsys_state is subsys's matter.
598 * Looking up and scanning function should be called under rcu_read_lock().
599 * Taking cgroup_mutex is not necessary for following calls.
600 * But the css returned by this routine can be "not populated yet" or "being
601 * destroyed". The caller should check css and cgroup's status.
605 * Typically Called at ->destroy(), or somewhere the subsys frees
606 * cgroup_subsys_state.
608 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
610 /* Find a cgroup_subsys_state which has given ID */
612 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
615 * Get a cgroup whose id is greater than or equal to id under tree of root.
616 * Returning a cgroup_subsys_state or NULL.
618 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
619 struct cgroup_subsys_state *root, int *foundid);
621 /* Returns true if root is ancestor of cg */
622 bool css_is_ancestor(struct cgroup_subsys_state *cg,
623 const struct cgroup_subsys_state *root);
625 /* Get id and depth of css */
626 unsigned short css_id(struct cgroup_subsys_state *css);
627 unsigned short css_depth(struct cgroup_subsys_state *css);
628 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
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)
654 #endif /* !CONFIG_CGROUPS */
656 #endif /* _LINUX_CGROUP_H */