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>
29 extern int cgroup_init_early(void);
30 extern int cgroup_init(void);
31 extern void cgroup_lock(void);
32 extern int cgroup_lock_is_held(void);
33 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
34 extern void cgroup_unlock(void);
35 extern void cgroup_fork(struct task_struct *p);
36 extern void cgroup_fork_callbacks(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 enum cgroup_subsys_id {
49 #include <linux/cgroup_subsys.h>
50 CGROUP_BUILTIN_SUBSYS_COUNT
54 * This define indicates the maximum number of subsystems that can be loaded
55 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
56 * track of all of them.
58 #define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
60 /* Per-subsystem/per-cgroup state maintained by the system. */
61 struct cgroup_subsys_state {
63 * The cgroup that this subsystem is attached to. Useful
64 * for subsystems that want to know about the cgroup
67 struct cgroup *cgroup;
70 * State maintained by the cgroup system to allow subsystems
71 * to be "busy". Should be accessed via css_get(),
72 * css_tryget() and and css_put().
78 /* ID for this css, if possible */
79 struct css_id __rcu *id;
81 /* Used to put @cgroup->dentry on the last css_put() */
82 struct work_struct dput_work;
85 /* bits in struct cgroup_subsys_state flags field */
87 CSS_ROOT, /* This CSS is the root of the subsystem */
90 /* Caller must verify that the css is not for root cgroup */
91 static inline void __css_get(struct cgroup_subsys_state *css, int count)
93 atomic_add(count, &css->refcnt);
97 * Call css_get() to hold a reference on the css; it can be used
98 * for a reference obtained via:
99 * - an existing ref-counted reference to the css
100 * - task->cgroups for a locked task
103 static inline void css_get(struct cgroup_subsys_state *css)
105 /* We don't need to reference count the root state */
106 if (!test_bit(CSS_ROOT, &css->flags))
111 * Call css_tryget() to take a reference on a css if your existing
112 * (known-valid) reference isn't already ref-counted. Returns false if
113 * the css has been destroyed.
116 extern bool __css_tryget(struct cgroup_subsys_state *css);
117 static inline bool css_tryget(struct cgroup_subsys_state *css)
119 if (test_bit(CSS_ROOT, &css->flags))
121 return __css_tryget(css);
125 * css_put() should be called to release a reference taken by
126 * css_get() or css_tryget()
129 extern void __css_put(struct cgroup_subsys_state *css);
130 static inline void css_put(struct cgroup_subsys_state *css)
132 if (!test_bit(CSS_ROOT, &css->flags))
136 /* bits in struct cgroup flags field */
138 /* Control Group is dead */
141 * Control Group has previously had a child cgroup or a task,
142 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
145 /* Control Group requires release notifications to userspace */
146 CGRP_NOTIFY_ON_RELEASE,
148 * A thread in rmdir() is wating for this cgroup.
152 * Clone cgroup values when creating a new child cgroup
158 unsigned long flags; /* "unsigned long" so bitops work */
161 * count users of this cgroup. >0 means busy, but doesn't
162 * necessarily indicate the number of tasks in the cgroup
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 __rcu *dentry; /* cgroup fs entry, RCU protected */
177 /* Private pointers for each registered subsystem */
178 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
180 struct cgroupfs_root *root;
181 struct cgroup *top_cgroup;
184 * List of cg_cgroup_links pointing at css_sets with
185 * tasks in this cgroup. Protected by css_set_lock
187 struct list_head css_sets;
189 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
190 struct list_head cft_q_node; /* used during cftype add/rm */
193 * Linked list running through all cgroups that can
194 * potentially be reaped by the release agent. Protected by
197 struct list_head release_list;
200 * list of pidlists, up to two for each namespace (one for procs, one
201 * for tasks); created on demand.
203 struct list_head pidlists;
204 struct mutex pidlist_mutex;
206 /* For RCU-protected deletion */
207 struct rcu_head rcu_head;
209 /* List of events which userspace want to receive */
210 struct list_head event_list;
211 spinlock_t event_list_lock;
215 * A css_set is a structure holding pointers to a set of
216 * cgroup_subsys_state objects. This saves space in the task struct
217 * object and speeds up fork()/exit(), since a single inc/dec and a
218 * list_add()/del() can bump the reference count on the entire cgroup
224 /* Reference count */
228 * List running through all cgroup groups in the same hash
229 * slot. Protected by css_set_lock
231 struct hlist_node hlist;
234 * List running through all tasks using this cgroup
235 * group. Protected by css_set_lock
237 struct list_head tasks;
240 * List of cg_cgroup_link objects on link chains from
241 * cgroups referenced from this css_set. Protected by
244 struct list_head cg_links;
247 * Set of subsystem states, one for each subsystem. This array
248 * is immutable after creation apart from the init_css_set
249 * during subsystem registration (at boot time) and modular subsystem
252 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
254 /* For RCU-protected deletion */
255 struct rcu_head rcu_head;
259 * cgroup_map_cb is an abstract callback API for reporting map-valued
263 struct cgroup_map_cb {
264 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
269 * struct cftype: handler definitions for cgroup control files
271 * When reading/writing to a file:
272 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
273 * - the 'cftype' of the file is file->f_dentry->d_fsdata
277 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
278 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
280 #define MAX_CFTYPE_NAME 64
284 * By convention, the name should begin with the name of the
285 * subsystem, followed by a period. Zero length string indicates
286 * end of cftype array.
288 char name[MAX_CFTYPE_NAME];
291 * If not 0, file mode is set to this value, otherwise it will
292 * be figured out automatically
297 * If non-zero, defines the maximum length of string that can
298 * be passed to write_string; defaults to 64
300 size_t max_write_len;
305 int (*open)(struct inode *inode, struct file *file);
306 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
308 char __user *buf, size_t nbytes, loff_t *ppos);
310 * read_u64() is a shortcut for the common case of returning a
311 * single integer. Use it in place of read()
313 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
315 * read_s64() is a signed version of read_u64()
317 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
319 * read_map() is used for defining a map of key/value
320 * pairs. It should call cb->fill(cb, key, value) for each
321 * entry. The key/value pairs (and their ordering) should not
322 * change between reboots.
324 int (*read_map)(struct cgroup *cont, struct cftype *cft,
325 struct cgroup_map_cb *cb);
327 * read_seq_string() is used for outputting a simple sequence
330 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
333 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
335 const char __user *buf, size_t nbytes, loff_t *ppos);
338 * write_u64() is a shortcut for the common case of accepting
339 * a single integer (as parsed by simple_strtoull) from
340 * userspace. Use in place of write(); return 0 or error.
342 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
344 * write_s64() is a signed version of write_u64()
346 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
349 * write_string() is passed a nul-terminated kernelspace
350 * buffer of maximum length determined by max_write_len.
351 * Returns 0 or -ve error code.
353 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
356 * trigger() callback can be used to get some kick from the
357 * userspace, when the actual string written is not important
358 * at all. The private field can be used to determine the
359 * kick type for multiplexing.
361 int (*trigger)(struct cgroup *cgrp, unsigned int event);
363 int (*release)(struct inode *inode, struct file *file);
366 * register_event() callback will be used to add new userspace
367 * waiter for changes related to the cftype. Implement it if
368 * you want to provide this functionality. Use eventfd_signal()
369 * on eventfd to send notification to userspace.
371 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
372 struct eventfd_ctx *eventfd, const char *args);
374 * unregister_event() callback will be called when userspace
375 * closes the eventfd or on cgroup removing.
376 * This callback must be implemented, if you want provide
377 * notification functionality.
379 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
380 struct eventfd_ctx *eventfd);
384 * cftype_sets describe cftypes belonging to a subsystem and are chained at
385 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
386 * terminated by zero length name.
389 struct list_head node; /* chained at subsys->cftsets */
390 const struct cftype *cfts;
393 struct cgroup_scanner {
395 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
396 void (*process_task)(struct task_struct *p,
397 struct cgroup_scanner *scan);
398 struct ptr_heap *heap;
402 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts);
403 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts);
405 int cgroup_is_removed(const struct cgroup *cgrp);
407 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
409 int cgroup_task_count(const struct cgroup *cgrp);
411 /* Return true if cgrp is a descendant of the task's cgroup */
412 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
415 * When the subsys has to access css and may add permanent refcnt to css,
416 * it should take care of racy conditions with rmdir(). Following set of
417 * functions, is for stop/restart rmdir if necessary.
418 * Because these will call css_get/put, "css" should be alive css.
420 * cgroup_exclude_rmdir();
421 * ...do some jobs which may access arbitrary empty cgroup
422 * cgroup_release_and_wakeup_rmdir();
424 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
425 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
428 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
429 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
432 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
435 struct cgroup_taskset;
436 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
437 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
438 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
439 int cgroup_taskset_size(struct cgroup_taskset *tset);
442 * cgroup_taskset_for_each - iterate cgroup_taskset
443 * @task: the loop cursor
444 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
445 * @tset: taskset to iterate
447 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
448 for ((task) = cgroup_taskset_first((tset)); (task); \
449 (task) = cgroup_taskset_next((tset))) \
450 if (!(skip_cgrp) || \
451 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
454 * Control Group subsystem type.
455 * See Documentation/cgroups/cgroups.txt for details
458 struct cgroup_subsys {
459 struct cgroup_subsys_state *(*create)(struct cgroup *cgrp);
460 int (*pre_destroy)(struct cgroup *cgrp);
461 void (*destroy)(struct cgroup *cgrp);
462 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
463 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
464 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
465 void (*fork)(struct task_struct *task);
466 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
467 struct task_struct *task);
468 void (*post_clone)(struct cgroup *cgrp);
469 void (*bind)(struct cgroup *root);
476 * True if this subsys uses ID. ID is not available before cgroup_init()
477 * (not available in early_init time.)
481 #define MAX_CGROUP_TYPE_NAMELEN 32
485 * Link to parent, and list entry in parent's children.
486 * Protected by cgroup_lock()
488 struct cgroupfs_root *root;
489 struct list_head sibling;
490 /* used when use_id == true */
494 /* list of cftype_sets */
495 struct list_head cftsets;
497 /* base cftypes, automatically [de]registered with subsys itself */
498 struct cftype *base_cftypes;
499 struct cftype_set base_cftset;
501 /* should be defined only by modular subsystems */
502 struct module *module;
505 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
506 #include <linux/cgroup_subsys.h>
509 static inline struct cgroup_subsys_state *cgroup_subsys_state(
510 struct cgroup *cgrp, int subsys_id)
512 return cgrp->subsys[subsys_id];
516 * function to get the cgroup_subsys_state which allows for extra
517 * rcu_dereference_check() conditions, such as locks used during the
518 * cgroup_subsys::attach() methods.
520 #define task_subsys_state_check(task, subsys_id, __c) \
521 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
522 lockdep_is_held(&task->alloc_lock) || \
523 cgroup_lock_is_held() || (__c))
525 static inline struct cgroup_subsys_state *
526 task_subsys_state(struct task_struct *task, int subsys_id)
528 return task_subsys_state_check(task, subsys_id, false);
531 static inline struct cgroup* task_cgroup(struct task_struct *task,
534 return task_subsys_state(task, subsys_id)->cgroup;
537 /* A cgroup_iter should be treated as an opaque object */
539 struct list_head *cg_link;
540 struct list_head *task;
544 * To iterate across the tasks in a cgroup:
546 * 1) call cgroup_iter_start to initialize an iterator
548 * 2) call cgroup_iter_next() to retrieve member tasks until it
549 * returns NULL or until you want to end the iteration
551 * 3) call cgroup_iter_end() to destroy the iterator.
553 * Or, call cgroup_scan_tasks() to iterate through every task in a
554 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
555 * the test_task() callback, but not while calling the process_task()
558 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
559 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
560 struct cgroup_iter *it);
561 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
562 int cgroup_scan_tasks(struct cgroup_scanner *scan);
563 int cgroup_attach_task(struct cgroup *, struct task_struct *);
564 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
567 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
568 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
569 * CSS ID is assigned at cgroup allocation (create) automatically
570 * and removed when subsys calls free_css_id() function. This is because
571 * the lifetime of cgroup_subsys_state is subsys's matter.
573 * Looking up and scanning function should be called under rcu_read_lock().
574 * Taking cgroup_mutex is not necessary for following calls.
575 * But the css returned by this routine can be "not populated yet" or "being
576 * destroyed". The caller should check css and cgroup's status.
580 * Typically Called at ->destroy(), or somewhere the subsys frees
581 * cgroup_subsys_state.
583 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
585 /* Find a cgroup_subsys_state which has given ID */
587 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
590 * Get a cgroup whose id is greater than or equal to id under tree of root.
591 * Returning a cgroup_subsys_state or NULL.
593 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
594 struct cgroup_subsys_state *root, int *foundid);
596 /* Returns true if root is ancestor of cg */
597 bool css_is_ancestor(struct cgroup_subsys_state *cg,
598 const struct cgroup_subsys_state *root);
600 /* Get id and depth of css */
601 unsigned short css_id(struct cgroup_subsys_state *css);
602 unsigned short css_depth(struct cgroup_subsys_state *css);
603 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
605 #else /* !CONFIG_CGROUPS */
607 static inline int cgroup_init_early(void) { return 0; }
608 static inline int cgroup_init(void) { return 0; }
609 static inline void cgroup_fork(struct task_struct *p) {}
610 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
611 static inline void cgroup_post_fork(struct task_struct *p) {}
612 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
614 static inline void cgroup_lock(void) {}
615 static inline void cgroup_unlock(void) {}
616 static inline int cgroupstats_build(struct cgroupstats *stats,
617 struct dentry *dentry)
622 /* No cgroups - nothing to do */
623 static inline int cgroup_attach_task_all(struct task_struct *from,
624 struct task_struct *t)
629 #endif /* !CONFIG_CGROUPS */
631 #endif /* _LINUX_CGROUP_H */