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>
26 extern int cgroup_init_early(void);
27 extern int cgroup_init(void);
28 extern void cgroup_lock(void);
29 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
30 extern void cgroup_unlock(void);
31 extern void cgroup_fork(struct task_struct *p);
32 extern void cgroup_fork_callbacks(struct task_struct *p);
33 extern void cgroup_post_fork(struct task_struct *p);
34 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
35 extern int cgroupstats_build(struct cgroupstats *stats,
36 struct dentry *dentry);
38 extern struct file_operations proc_cgroup_operations;
40 /* Define the enumeration of all cgroup subsystems */
41 #define SUBSYS(_x) _x ## _subsys_id,
42 enum cgroup_subsys_id {
43 #include <linux/cgroup_subsys.h>
48 /* Per-subsystem/per-cgroup state maintained by the system. */
49 struct cgroup_subsys_state {
50 /* The cgroup that this subsystem is attached to. Useful
51 * for subsystems that want to know about the cgroup
52 * hierarchy structure */
53 struct cgroup *cgroup;
55 /* State maintained by the cgroup system to allow subsystems
56 * to be "busy". Should be accessed via css_get(),
57 * css_tryget() and and css_put(). */
64 /* bits in struct cgroup_subsys_state flags field */
66 CSS_ROOT, /* This CSS is the root of the subsystem */
67 CSS_REMOVED, /* This CSS is dead */
71 * Call css_get() to hold a reference on the css; it can be used
72 * for a reference obtained via:
73 * - an existing ref-counted reference to the css
74 * - task->cgroups for a locked task
77 static inline void css_get(struct cgroup_subsys_state *css)
79 /* We don't need to reference count the root state */
80 if (!test_bit(CSS_ROOT, &css->flags))
81 atomic_inc(&css->refcnt);
84 static inline bool css_is_removed(struct cgroup_subsys_state *css)
86 return test_bit(CSS_REMOVED, &css->flags);
90 * Call css_tryget() to take a reference on a css if your existing
91 * (known-valid) reference isn't already ref-counted. Returns false if
92 * the css has been destroyed.
95 static inline bool css_tryget(struct cgroup_subsys_state *css)
97 if (test_bit(CSS_ROOT, &css->flags))
99 while (!atomic_inc_not_zero(&css->refcnt)) {
100 if (test_bit(CSS_REMOVED, &css->flags))
107 * css_put() should be called to release a reference taken by
108 * css_get() or css_tryget()
111 extern void __css_put(struct cgroup_subsys_state *css);
112 static inline void css_put(struct cgroup_subsys_state *css)
114 if (!test_bit(CSS_ROOT, &css->flags))
118 /* bits in struct cgroup flags field */
120 /* Control Group is dead */
122 /* Control Group has previously had a child cgroup or a task,
123 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
125 /* Control Group requires release notifications to userspace */
126 CGRP_NOTIFY_ON_RELEASE,
130 unsigned long flags; /* "unsigned long" so bitops work */
132 /* count users of this cgroup. >0 means busy, but doesn't
133 * necessarily indicate the number of tasks in the
138 * We link our 'sibling' struct into our parent's 'children'.
139 * Our children link their 'sibling' into our 'children'.
141 struct list_head sibling; /* my parent's children */
142 struct list_head children; /* my children */
144 struct cgroup *parent; /* my parent */
145 struct dentry *dentry; /* cgroup fs entry, RCU protected */
147 /* Private pointers for each registered subsystem */
148 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
150 struct cgroupfs_root *root;
151 struct cgroup *top_cgroup;
154 * List of cg_cgroup_links pointing at css_sets with
155 * tasks in this cgroup. Protected by css_set_lock
157 struct list_head css_sets;
160 * Linked list running through all cgroups that can
161 * potentially be reaped by the release agent. Protected by
164 struct list_head release_list;
166 /* pids_mutex protects the fields below */
167 struct rw_semaphore pids_mutex;
168 /* Array of process ids in the cgroup */
170 /* How many files are using the current tasks_pids array */
172 /* Length of the current tasks_pids array */
175 /* For RCU-protected deletion */
176 struct rcu_head rcu_head;
179 /* A css_set is a structure holding pointers to a set of
180 * cgroup_subsys_state objects. This saves space in the task struct
181 * object and speeds up fork()/exit(), since a single inc/dec and a
182 * list_add()/del() can bump the reference count on the entire
183 * cgroup set for a task.
188 /* Reference count */
192 * List running through all cgroup groups in the same hash
193 * slot. Protected by css_set_lock
195 struct hlist_node hlist;
198 * List running through all tasks using this cgroup
199 * group. Protected by css_set_lock
201 struct list_head tasks;
204 * List of cg_cgroup_link objects on link chains from
205 * cgroups referenced from this css_set. Protected by
208 struct list_head cg_links;
211 * Set of subsystem states, one for each subsystem. This array
212 * is immutable after creation apart from the init_css_set
213 * during subsystem registration (at boot time).
215 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
219 * cgroup_map_cb is an abstract callback API for reporting map-valued
223 struct cgroup_map_cb {
224 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
230 * The files in the cgroup filesystem mostly have a very simple read/write
231 * handling, some common function will take care of it. Nevertheless some cases
232 * (read tasks) are special and therefore I define this structure for every
236 * When reading/writing to a file:
237 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
238 * - the 'cftype' of the file is file->f_dentry->d_fsdata
241 #define MAX_CFTYPE_NAME 64
243 /* By convention, the name should begin with the name of the
244 * subsystem, followed by a period */
245 char name[MAX_CFTYPE_NAME];
249 * If non-zero, defines the maximum length of string that can
250 * be passed to write_string; defaults to 64
252 size_t max_write_len;
254 int (*open)(struct inode *inode, struct file *file);
255 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
257 char __user *buf, size_t nbytes, loff_t *ppos);
259 * read_u64() is a shortcut for the common case of returning a
260 * single integer. Use it in place of read()
262 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
264 * read_s64() is a signed version of read_u64()
266 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
268 * read_map() is used for defining a map of key/value
269 * pairs. It should call cb->fill(cb, key, value) for each
270 * entry. The key/value pairs (and their ordering) should not
271 * change between reboots.
273 int (*read_map)(struct cgroup *cont, struct cftype *cft,
274 struct cgroup_map_cb *cb);
276 * read_seq_string() is used for outputting a simple sequence
279 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
282 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
284 const char __user *buf, size_t nbytes, loff_t *ppos);
287 * write_u64() is a shortcut for the common case of accepting
288 * a single integer (as parsed by simple_strtoull) from
289 * userspace. Use in place of write(); return 0 or error.
291 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
293 * write_s64() is a signed version of write_u64()
295 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
298 * write_string() is passed a nul-terminated kernelspace
299 * buffer of maximum length determined by max_write_len.
300 * Returns 0 or -ve error code.
302 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
305 * trigger() callback can be used to get some kick from the
306 * userspace, when the actual string written is not important
307 * at all. The private field can be used to determine the
308 * kick type for multiplexing.
310 int (*trigger)(struct cgroup *cgrp, unsigned int event);
312 int (*release)(struct inode *inode, struct file *file);
315 struct cgroup_scanner {
317 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
318 void (*process_task)(struct task_struct *p,
319 struct cgroup_scanner *scan);
320 struct ptr_heap *heap;
323 /* Add a new file to the given cgroup directory. Should only be
324 * called by subsystems from within a populate() method */
325 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
326 const struct cftype *cft);
328 /* Add a set of new files to the given cgroup directory. Should
329 * only be called by subsystems from within a populate() method */
330 int cgroup_add_files(struct cgroup *cgrp,
331 struct cgroup_subsys *subsys,
332 const struct cftype cft[],
335 int cgroup_is_removed(const struct cgroup *cgrp);
337 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
339 int cgroup_task_count(const struct cgroup *cgrp);
341 /* Return true if the cgroup is a descendant of the current cgroup */
342 int cgroup_is_descendant(const struct cgroup *cgrp);
344 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
346 struct cgroup_subsys {
347 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
348 struct cgroup *cgrp);
349 void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
350 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
351 int (*can_attach)(struct cgroup_subsys *ss,
352 struct cgroup *cgrp, struct task_struct *tsk);
353 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
354 struct cgroup *old_cgrp, struct task_struct *tsk);
355 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
356 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
357 int (*populate)(struct cgroup_subsys *ss,
358 struct cgroup *cgrp);
359 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
360 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
366 #define MAX_CGROUP_TYPE_NAMELEN 32
370 * Protects sibling/children links of cgroups in this
371 * hierarchy, plus protects which hierarchy (or none) the
372 * subsystem is a part of (i.e. root/sibling). To avoid
373 * potential deadlocks, the following operations should not be
374 * undertaken while holding any hierarchy_mutex:
376 * - allocating memory
377 * - initiating hotplug events
379 struct mutex hierarchy_mutex;
382 * Link to parent, and list entry in parent's children.
383 * Protected by this->hierarchy_mutex and cgroup_lock()
385 struct cgroupfs_root *root;
386 struct list_head sibling;
389 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
390 #include <linux/cgroup_subsys.h>
393 static inline struct cgroup_subsys_state *cgroup_subsys_state(
394 struct cgroup *cgrp, int subsys_id)
396 return cgrp->subsys[subsys_id];
399 static inline struct cgroup_subsys_state *task_subsys_state(
400 struct task_struct *task, int subsys_id)
402 return rcu_dereference(task->cgroups->subsys[subsys_id]);
405 static inline struct cgroup* task_cgroup(struct task_struct *task,
408 return task_subsys_state(task, subsys_id)->cgroup;
411 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
414 /* A cgroup_iter should be treated as an opaque object */
416 struct list_head *cg_link;
417 struct list_head *task;
420 /* To iterate across the tasks in a cgroup:
422 * 1) call cgroup_iter_start to intialize an iterator
424 * 2) call cgroup_iter_next() to retrieve member tasks until it
425 * returns NULL or until you want to end the iteration
427 * 3) call cgroup_iter_end() to destroy the iterator.
429 * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
430 * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
431 * callback, but not while calling the process_task() callback.
433 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
434 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
435 struct cgroup_iter *it);
436 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
437 int cgroup_scan_tasks(struct cgroup_scanner *scan);
438 int cgroup_attach_task(struct cgroup *, struct task_struct *);
440 #else /* !CONFIG_CGROUPS */
442 static inline int cgroup_init_early(void) { return 0; }
443 static inline int cgroup_init(void) { return 0; }
444 static inline void cgroup_fork(struct task_struct *p) {}
445 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
446 static inline void cgroup_post_fork(struct task_struct *p) {}
447 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
449 static inline void cgroup_lock(void) {}
450 static inline void cgroup_unlock(void) {}
451 static inline int cgroupstats_build(struct cgroupstats *stats,
452 struct dentry *dentry)
457 #endif /* !CONFIG_CGROUPS */
459 #endif /* _LINUX_CGROUP_H */