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 in the same hash
154 * slot. Protected by css_set_lock
156 struct hlist_node hlist
;
159 * List running through all tasks using this cgroup
160 * group. Protected by css_set_lock
162 struct list_head tasks
;
165 * List of cg_cgroup_link objects on link chains from
166 * cgroups referenced from this css_set. Protected by
169 struct list_head cg_links
;
172 * Set of subsystem states, one for each subsystem. This array
173 * is immutable after creation apart from the init_css_set
174 * during subsystem registration (at boot time).
176 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
180 * cgroup_map_cb is an abstract callback API for reporting map-valued
184 struct cgroup_map_cb
{
185 int (*fill
)(struct cgroup_map_cb
*cb
, const char *key
, u64 value
);
191 * The files in the cgroup filesystem mostly have a very simple read/write
192 * handling, some common function will take care of it. Nevertheless some cases
193 * (read tasks) are special and therefore I define this structure for every
197 * When reading/writing to a file:
198 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
199 * - the 'cftype' of the file is file->f_dentry->d_fsdata
202 #define MAX_CFTYPE_NAME 64
204 /* By convention, the name should begin with the name of the
205 * subsystem, followed by a period */
206 char name
[MAX_CFTYPE_NAME
];
210 * If non-zero, defines the maximum length of string that can
211 * be passed to write_string; defaults to 64
213 size_t max_write_len
;
215 int (*open
)(struct inode
*inode
, struct file
*file
);
216 ssize_t (*read
)(struct cgroup
*cgrp
, struct cftype
*cft
,
218 char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
220 * read_u64() is a shortcut for the common case of returning a
221 * single integer. Use it in place of read()
223 u64 (*read_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
225 * read_s64() is a signed version of read_u64()
227 s64 (*read_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
);
229 * read_map() is used for defining a map of key/value
230 * pairs. It should call cb->fill(cb, key, value) for each
231 * entry. The key/value pairs (and their ordering) should not
232 * change between reboots.
234 int (*read_map
)(struct cgroup
*cont
, struct cftype
*cft
,
235 struct cgroup_map_cb
*cb
);
237 * read_seq_string() is used for outputting a simple sequence
240 int (*read_seq_string
)(struct cgroup
*cont
, struct cftype
*cft
,
243 ssize_t (*write
)(struct cgroup
*cgrp
, struct cftype
*cft
,
245 const char __user
*buf
, size_t nbytes
, loff_t
*ppos
);
248 * write_u64() is a shortcut for the common case of accepting
249 * a single integer (as parsed by simple_strtoull) from
250 * userspace. Use in place of write(); return 0 or error.
252 int (*write_u64
)(struct cgroup
*cgrp
, struct cftype
*cft
, u64 val
);
254 * write_s64() is a signed version of write_u64()
256 int (*write_s64
)(struct cgroup
*cgrp
, struct cftype
*cft
, s64 val
);
259 * write_string() is passed a nul-terminated kernelspace
260 * buffer of maximum length determined by max_write_len.
261 * Returns 0 or -ve error code.
263 int (*write_string
)(struct cgroup
*cgrp
, struct cftype
*cft
,
266 * trigger() callback can be used to get some kick from the
267 * userspace, when the actual string written is not important
268 * at all. The private field can be used to determine the
269 * kick type for multiplexing.
271 int (*trigger
)(struct cgroup
*cgrp
, unsigned int event
);
273 int (*release
)(struct inode
*inode
, struct file
*file
);
276 struct cgroup_scanner
{
278 int (*test_task
)(struct task_struct
*p
, struct cgroup_scanner
*scan
);
279 void (*process_task
)(struct task_struct
*p
,
280 struct cgroup_scanner
*scan
);
281 struct ptr_heap
*heap
;
284 /* Add a new file to the given cgroup directory. Should only be
285 * called by subsystems from within a populate() method */
286 int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
287 const struct cftype
*cft
);
289 /* Add a set of new files to the given cgroup directory. Should
290 * only be called by subsystems from within a populate() method */
291 int cgroup_add_files(struct cgroup
*cgrp
,
292 struct cgroup_subsys
*subsys
,
293 const struct cftype cft
[],
296 int cgroup_is_removed(const struct cgroup
*cgrp
);
298 int cgroup_lock_live_group(struct cgroup
*cgrp
);
300 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
);
302 int cgroup_task_count(const struct cgroup
*cgrp
);
304 /* Return true if the cgroup is a descendant of the current cgroup */
305 int cgroup_is_descendant(const struct cgroup
*cgrp
);
307 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
309 struct cgroup_subsys
{
310 struct cgroup_subsys_state
*(*create
)(struct cgroup_subsys
*ss
,
311 struct cgroup
*cgrp
);
312 void (*pre_destroy
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
313 void (*destroy
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
314 int (*can_attach
)(struct cgroup_subsys
*ss
,
315 struct cgroup
*cgrp
, struct task_struct
*tsk
);
316 void (*attach
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
,
317 struct cgroup
*old_cgrp
, struct task_struct
*tsk
);
318 void (*fork
)(struct cgroup_subsys
*ss
, struct task_struct
*task
);
319 void (*exit
)(struct cgroup_subsys
*ss
, struct task_struct
*task
);
320 int (*populate
)(struct cgroup_subsys
*ss
,
321 struct cgroup
*cgrp
);
322 void (*post_clone
)(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
);
323 void (*bind
)(struct cgroup_subsys
*ss
, struct cgroup
*root
);
325 * This routine is called with the task_lock of mm->owner held
327 void (*mm_owner_changed
)(struct cgroup_subsys
*ss
,
334 #define MAX_CGROUP_TYPE_NAMELEN 32
337 /* Protected by RCU */
338 struct cgroupfs_root
*root
;
340 struct list_head sibling
;
345 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
346 #include <linux/cgroup_subsys.h>
349 static inline struct cgroup_subsys_state
*cgroup_subsys_state(
350 struct cgroup
*cgrp
, int subsys_id
)
352 return cgrp
->subsys
[subsys_id
];
355 static inline struct cgroup_subsys_state
*task_subsys_state(
356 struct task_struct
*task
, int subsys_id
)
358 return rcu_dereference(task
->cgroups
->subsys
[subsys_id
]);
361 static inline struct cgroup
* task_cgroup(struct task_struct
*task
,
364 return task_subsys_state(task
, subsys_id
)->cgroup
;
367 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*ss
);
369 /* A cgroup_iter should be treated as an opaque object */
371 struct list_head
*cg_link
;
372 struct list_head
*task
;
375 /* To iterate across the tasks in a cgroup:
377 * 1) call cgroup_iter_start to intialize an iterator
379 * 2) call cgroup_iter_next() to retrieve member tasks until it
380 * returns NULL or until you want to end the iteration
382 * 3) call cgroup_iter_end() to destroy the iterator.
384 * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
385 * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
386 * callback, but not while calling the process_task() callback.
388 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
389 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
390 struct cgroup_iter
*it
);
391 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
);
392 int cgroup_scan_tasks(struct cgroup_scanner
*scan
);
393 int cgroup_attach_task(struct cgroup
*, struct task_struct
*);
395 #else /* !CONFIG_CGROUPS */
397 static inline int cgroup_init_early(void) { return 0; }
398 static inline int cgroup_init(void) { return 0; }
399 static inline void cgroup_init_smp(void) {}
400 static inline void cgroup_fork(struct task_struct
*p
) {}
401 static inline void cgroup_fork_callbacks(struct task_struct
*p
) {}
402 static inline void cgroup_post_fork(struct task_struct
*p
) {}
403 static inline void cgroup_exit(struct task_struct
*p
, int callbacks
) {}
405 static inline void cgroup_lock(void) {}
406 static inline void cgroup_unlock(void) {}
407 static inline int cgroupstats_build(struct cgroupstats
*stats
,
408 struct dentry
*dentry
)
413 #endif /* !CONFIG_CGROUPS */
415 #ifdef CONFIG_MM_OWNER
417 cgroup_mm_owner_callbacks(struct task_struct
*old
, struct task_struct
*new);
418 #else /* !CONFIG_MM_OWNER */
420 cgroup_mm_owner_callbacks(struct task_struct
*old
, struct task_struct
*new)
423 #endif /* CONFIG_MM_OWNER */
424 #endif /* _LINUX_CGROUP_H */