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/kref.h>
13 #include <linux/cpumask.h>
14 #include <linux/nodemask.h>
15 #include <linux/rcupdate.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/prio_heap.h>
25 extern int cgroup_init_early(void);
26 extern int cgroup_init(void);
27 extern void cgroup_init_smp(void);
28 extern void cgroup_lock(void);
29 extern void cgroup_unlock(void);
30 extern void cgroup_fork(struct task_struct
*p
);
31 extern void cgroup_fork_callbacks(struct task_struct
*p
);
32 extern void cgroup_post_fork(struct task_struct
*p
);
33 extern void cgroup_exit(struct task_struct
*p
, int run_callbacks
);
34 extern int cgroupstats_build(struct cgroupstats
*stats
,
35 struct dentry
*dentry
);
37 extern struct file_operations proc_cgroup_operations
;
39 /* Define the enumeration of all cgroup subsystems */
40 #define SUBSYS(_x) _x ## _subsys_id,
41 enum cgroup_subsys_id
{
42 #include <linux/cgroup_subsys.h>
47 /* Per-subsystem/per-cgroup state maintained by the system. */
48 struct cgroup_subsys_state
{
49 /* The cgroup that this subsystem is attached to. Useful
50 * for subsystems that want to know about the cgroup
51 * hierarchy structure */
52 struct cgroup
*cgroup
;
54 /* State maintained by the cgroup system to allow
55 * subsystems to be "busy". Should be accessed via css_get()
63 /* bits in struct cgroup_subsys_state flags field */
65 CSS_ROOT
, /* This CSS is the root of the subsystem */
69 * Call css_get() to hold a reference on the cgroup;
73 static inline void css_get(struct cgroup_subsys_state
*css
)
75 /* We don't need to reference count the root state */
76 if (!test_bit(CSS_ROOT
, &css
->flags
))
77 atomic_inc(&css
->refcnt
);
80 * css_put() should be called to release a reference taken by
84 extern void __css_put(struct cgroup_subsys_state
*css
);
85 static inline void css_put(struct cgroup_subsys_state
*css
)
87 if (!test_bit(CSS_ROOT
, &css
->flags
))
91 /* bits in struct cgroup flags field */
93 /* Control Group is dead */
95 /* Control Group has previously had a child cgroup or a task,
96 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
98 /* Control Group requires release notifications to userspace */
99 CGRP_NOTIFY_ON_RELEASE
,
103 unsigned long flags
; /* "unsigned long" so bitops work */
105 /* count users of this cgroup. >0 means busy, but doesn't
106 * necessarily indicate the number of tasks in the
111 * We link our 'sibling' struct into our parent's 'children'.
112 * Our children link their 'sibling' into our 'children'.
114 struct list_head sibling
; /* my parent's children */
115 struct list_head children
; /* my children */
117 struct cgroup
*parent
; /* my parent */
118 struct dentry
*dentry
; /* cgroup fs entry */
120 /* Private pointers for each registered subsystem */
121 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
123 struct cgroupfs_root
*root
;
124 struct cgroup
*top_cgroup
;
127 * List of cg_cgroup_links pointing at css_sets with
128 * tasks in this cgroup. Protected by css_set_lock
130 struct list_head css_sets
;
133 * Linked list running through all cgroups that can
134 * potentially be reaped by the release agent. Protected by
137 struct list_head release_list
;
140 /* A css_set is a structure holding pointers to a set of
141 * cgroup_subsys_state objects. This saves space in the task struct
142 * object and speeds up fork()/exit(), since a single inc/dec and a
143 * list_add()/del() can bump the reference count on the entire
144 * cgroup set for a task.
149 /* Reference count */
153 * List running through all cgroup groups. Protected by
156 struct list_head list
;
159 * List running through all cgroup groups in the same hash
160 * slot. Protected by css_set_lock
162 struct hlist_node hlist
;
165 * List running through all tasks using this cgroup
166 * group. Protected by css_set_lock
168 struct list_head tasks
;
171 * List of cg_cgroup_link objects on link chains from
172 * cgroups referenced from this css_set. Protected by
175 struct list_head cg_links
;
178 * Set of subsystem states, one for each subsystem. This array
179 * is immutable after creation apart from the init_css_set
180 * during subsystem registration (at boot time).
182 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
186 * cgroup_map_cb is an abstract callback API for reporting map-valued
190 struct cgroup_map_cb
{
191 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
197 * The files in the cgroup filesystem mostly have a very simple read/write
198 * handling, some common function will take care of it. Nevertheless some cases
199 * (read tasks) are special and therefore I define this structure for every
203 * When reading/writing to a file:
204 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
205 * - the 'cftype' of the file is file->f_dentry->d_fsdata
208 #define MAX_CFTYPE_NAME 64
210 /* By convention, the name should begin with the name of the
211 * subsystem, followed by a period */
212 char name
[MAX_CFTYPE_NAME
];
214 int (*open
) (struct inode
*inode
, struct file
*file
);
215 ssize_t (*read
) (struct cgroup
*cgrp
, struct cftype
*cft
,
217 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
219 * read_u64() is a shortcut for the common case of returning a
220 * single integer. Use it in place of read()
222 u64 (*read_u64
) (struct cgroup
*cgrp
, struct cftype
*cft
);
224 * read_s64() is a signed version of read_u64()
226 s64 (*read_s64
) (struct cgroup
*cgrp
, struct cftype
*cft
);
228 * read_map() is used for defining a map of key/value
229 * pairs. It should call cb->fill(cb, key, value) for each
230 * entry. The key/value pairs (and their ordering) should not
231 * change between reboots.
233 int (*read_map
) (struct cgroup
*cont
, struct cftype
*cft
,
234 struct cgroup_map_cb
*cb
);
236 ssize_t (*write
) (struct cgroup
*cgrp
, struct cftype
*cft
,
238 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
241 * write_u64() is a shortcut for the common case of accepting
242 * a single integer (as parsed by simple_strtoull) from
243 * userspace. Use in place of write(); return 0 or error.
245 int (*write_u64
) (struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
247 * write_s64() is a signed version of write_u64()
249 int (*write_s64
) (struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
252 * trigger() callback can be used to get some kick from the
253 * userspace, when the actual string written is not important
254 * at all. The private field can be used to determine the
255 * kick type for multiplexing.
257 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
259 int (*release
) (struct inode
*inode
, struct file
*file
);
262 struct cgroup_scanner
{
264 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
265 void (*process_task
)(struct task_struct
*p
,
266 struct cgroup_scanner
*scan
);
267 struct ptr_heap
*heap
;
270 /* Add a new file to the given cgroup directory. Should only be
271 * called by subsystems from within a populate() method */
272 int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
273 const struct cftype
*cft
);
275 /* Add a set of new files to the given cgroup directory. Should
276 * only be called by subsystems from within a populate() method */
277 int cgroup_add_files(struct cgroup
*cgrp
,
278 struct cgroup_subsys
*subsys
,
279 const struct cftype cft
[],
282 int cgroup_is_removed(const struct cgroup
*cgrp
);
284 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
286 int cgroup_task_count(const struct cgroup
*cgrp
);
288 /* Return true if the cgroup is a descendant of the current cgroup */
289 int cgroup_is_descendant(const struct cgroup
*cgrp
);
291 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
293 struct cgroup_subsys
{
294 struct cgroup_subsys_state
*(*create
)(struct cgroup_subsys
*ss
,
295 struct cgroup
*cgrp
);
296 void (*pre_destroy
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
297 void (*destroy
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
298 int (*can_attach
)(struct cgroup_subsys
*ss
,
299 struct cgroup
*cgrp
, struct task_struct
*tsk
);
300 void (*attach
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
,
301 struct cgroup
*old_cgrp
, struct task_struct
*tsk
);
302 void (*fork
)(struct cgroup_subsys
*ss
, struct task_struct
*task
);
303 void (*exit
)(struct cgroup_subsys
*ss
, struct task_struct
*task
);
304 int (*populate
)(struct cgroup_subsys
*ss
,
305 struct cgroup
*cgrp
);
306 void (*post_clone
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
307 void (*bind
)(struct cgroup_subsys
*ss
, struct cgroup
*root
);
312 #define MAX_CGROUP_TYPE_NAMELEN 32
315 /* Protected by RCU */
316 struct cgroupfs_root
*root
;
318 struct list_head sibling
;
323 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
324 #include <linux/cgroup_subsys.h>
327 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
328 struct cgroup
*cgrp
, int subsys_id
)
330 return cgrp
->subsys
[subsys_id
];
333 static inline struct cgroup_subsys_state
*task_subsys_state(
334 struct task_struct
*task
, int subsys_id
)
336 return rcu_dereference(task
->cgroups
->subsys
[subsys_id
]);
339 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
342 return task_subsys_state(task
, subsys_id
)->cgroup
;
345 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*ss
);
347 /* A cgroup_iter should be treated as an opaque object */
349 struct list_head
*cg_link
;
350 struct list_head
*task
;
353 /* To iterate across the tasks in a cgroup:
355 * 1) call cgroup_iter_start to intialize an iterator
357 * 2) call cgroup_iter_next() to retrieve member tasks until it
358 * returns NULL or until you want to end the iteration
360 * 3) call cgroup_iter_end() to destroy the iterator.
362 * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
363 * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
364 * callback, but not while calling the process_task() callback.
366 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
367 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
368 struct cgroup_iter
*it
);
369 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
370 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
371 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
373 #else /* !CONFIG_CGROUPS */
375 static inline int cgroup_init_early(void) { return 0; }
376 static inline int cgroup_init(void) { return 0; }
377 static inline void cgroup_init_smp(void) {}
378 static inline void cgroup_fork(struct task_struct
*p
) {}
379 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
380 static inline void cgroup_post_fork(struct task_struct
*p
) {}
381 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
383 static inline void cgroup_lock(void) {}
384 static inline void cgroup_unlock(void) {}
385 static inline int cgroupstats_build(struct cgroupstats
*stats
,
386 struct dentry
*dentry
)
391 #endif /* !CONFIG_CGROUPS */
393 #endif /* _LINUX_CGROUP_H */