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>
28 extern int cgroup_init_early(void);
29 extern int cgroup_init(void);
30 extern void cgroup_lock(void);
31 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
32 extern void cgroup_unlock(void);
33 extern void cgroup_fork(struct task_struct *p);
34 extern void cgroup_fork_callbacks(struct task_struct *p);
35 extern void cgroup_post_fork(struct task_struct *p);
36 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
37 extern int cgroupstats_build(struct cgroupstats *stats,
38 struct dentry *dentry);
40 extern struct file_operations proc_cgroup_operations;
42 /* Define the enumeration of all cgroup subsystems */
43 #define SUBSYS(_x) _x ## _subsys_id,
44 enum cgroup_subsys_id {
45 #include <linux/cgroup_subsys.h>
50 /* Per-subsystem/per-cgroup state maintained by the system. */
51 struct cgroup_subsys_state {
53 * The cgroup that this subsystem is attached to. Useful
54 * for subsystems that want to know about the cgroup
57 struct cgroup *cgroup;
60 * State maintained by the cgroup system to allow subsystems
61 * to be "busy". Should be accessed via css_get(),
62 * css_tryget() and and css_put().
68 /* ID for this css, if possible */
72 /* bits in struct cgroup_subsys_state flags field */
74 CSS_ROOT, /* This CSS is the root of the subsystem */
75 CSS_REMOVED, /* This CSS is dead */
79 * Call css_get() to hold a reference on the css; it can be used
80 * for a reference obtained via:
81 * - an existing ref-counted reference to the css
82 * - task->cgroups for a locked task
85 static inline void css_get(struct cgroup_subsys_state *css)
87 /* We don't need to reference count the root state */
88 if (!test_bit(CSS_ROOT, &css->flags))
89 atomic_inc(&css->refcnt);
92 static inline bool css_is_removed(struct cgroup_subsys_state *css)
94 return test_bit(CSS_REMOVED, &css->flags);
98 * Call css_tryget() to take a reference on a css if your existing
99 * (known-valid) reference isn't already ref-counted. Returns false if
100 * the css has been destroyed.
103 static inline bool css_tryget(struct cgroup_subsys_state *css)
105 if (test_bit(CSS_ROOT, &css->flags))
107 while (!atomic_inc_not_zero(&css->refcnt)) {
108 if (test_bit(CSS_REMOVED, &css->flags))
116 * css_put() should be called to release a reference taken by
117 * css_get() or css_tryget()
120 extern void __css_put(struct cgroup_subsys_state *css);
121 static inline void css_put(struct cgroup_subsys_state *css)
123 if (!test_bit(CSS_ROOT, &css->flags))
127 /* bits in struct cgroup flags field */
129 /* Control Group is dead */
132 * Control Group has previously had a child cgroup or a task,
133 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
136 /* Control Group requires release notifications to userspace */
137 CGRP_NOTIFY_ON_RELEASE,
139 * A thread in rmdir() is wating for this cgroup.
145 unsigned long flags; /* "unsigned long" so bitops work */
148 * count users of this cgroup. >0 means busy, but doesn't
149 * necessarily indicate the number of tasks in the cgroup
154 * We link our 'sibling' struct into our parent's 'children'.
155 * Our children link their 'sibling' into our 'children'.
157 struct list_head sibling; /* my parent's children */
158 struct list_head children; /* my children */
160 struct cgroup *parent; /* my parent */
161 struct dentry *dentry; /* cgroup fs entry, RCU protected */
163 /* Private pointers for each registered subsystem */
164 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
166 struct cgroupfs_root *root;
167 struct cgroup *top_cgroup;
170 * List of cg_cgroup_links pointing at css_sets with
171 * tasks in this cgroup. Protected by css_set_lock
173 struct list_head css_sets;
176 * Linked list running through all cgroups that can
177 * potentially be reaped by the release agent. Protected by
180 struct list_head release_list;
182 /* pids_mutex protects the fields below */
183 struct rw_semaphore pids_mutex;
184 /* Array of process ids in the cgroup */
186 /* How many files are using the current tasks_pids array */
188 /* Length of the current tasks_pids array */
191 /* For RCU-protected deletion */
192 struct rcu_head rcu_head;
196 * A css_set is a structure holding pointers to a set of
197 * cgroup_subsys_state objects. This saves space in the task struct
198 * object and speeds up fork()/exit(), since a single inc/dec and a
199 * list_add()/del() can bump the reference count on the entire cgroup
205 /* Reference count */
209 * List running through all cgroup groups in the same hash
210 * slot. Protected by css_set_lock
212 struct hlist_node hlist;
215 * List running through all tasks using this cgroup
216 * group. Protected by css_set_lock
218 struct list_head tasks;
221 * List of cg_cgroup_link objects on link chains from
222 * cgroups referenced from this css_set. Protected by
225 struct list_head cg_links;
228 * Set of subsystem states, one for each subsystem. This array
229 * is immutable after creation apart from the init_css_set
230 * during subsystem registration (at boot time).
232 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
236 * cgroup_map_cb is an abstract callback API for reporting map-valued
240 struct cgroup_map_cb {
241 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
246 * struct cftype: handler definitions for cgroup control files
248 * When reading/writing to a file:
249 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
250 * - the 'cftype' of the file is file->f_dentry->d_fsdata
253 #define MAX_CFTYPE_NAME 64
256 * By convention, the name should begin with the name of the
257 * subsystem, followed by a period
259 char name[MAX_CFTYPE_NAME];
262 * If not 0, file mode is set to this value, otherwise it will
263 * be figured out automatically
268 * If non-zero, defines the maximum length of string that can
269 * be passed to write_string; defaults to 64
271 size_t max_write_len;
273 int (*open)(struct inode *inode, struct file *file);
274 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
276 char __user *buf, size_t nbytes, loff_t *ppos);
278 * read_u64() is a shortcut for the common case of returning a
279 * single integer. Use it in place of read()
281 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
283 * read_s64() is a signed version of read_u64()
285 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
287 * read_map() is used for defining a map of key/value
288 * pairs. It should call cb->fill(cb, key, value) for each
289 * entry. The key/value pairs (and their ordering) should not
290 * change between reboots.
292 int (*read_map)(struct cgroup *cont, struct cftype *cft,
293 struct cgroup_map_cb *cb);
295 * read_seq_string() is used for outputting a simple sequence
298 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
301 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
303 const char __user *buf, size_t nbytes, loff_t *ppos);
306 * write_u64() is a shortcut for the common case of accepting
307 * a single integer (as parsed by simple_strtoull) from
308 * userspace. Use in place of write(); return 0 or error.
310 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
312 * write_s64() is a signed version of write_u64()
314 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
317 * write_string() is passed a nul-terminated kernelspace
318 * buffer of maximum length determined by max_write_len.
319 * Returns 0 or -ve error code.
321 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
324 * trigger() callback can be used to get some kick from the
325 * userspace, when the actual string written is not important
326 * at all. The private field can be used to determine the
327 * kick type for multiplexing.
329 int (*trigger)(struct cgroup *cgrp, unsigned int event);
331 int (*release)(struct inode *inode, struct file *file);
334 struct cgroup_scanner {
336 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
337 void (*process_task)(struct task_struct *p,
338 struct cgroup_scanner *scan);
339 struct ptr_heap *heap;
344 * Add a new file to the given cgroup directory. Should only be
345 * called by subsystems from within a populate() method
347 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
348 const struct cftype *cft);
351 * Add a set of new files to the given cgroup directory. Should
352 * only be called by subsystems from within a populate() method
354 int cgroup_add_files(struct cgroup *cgrp,
355 struct cgroup_subsys *subsys,
356 const struct cftype cft[],
359 int cgroup_is_removed(const struct cgroup *cgrp);
361 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
363 int cgroup_task_count(const struct cgroup *cgrp);
365 /* Return true if cgrp is a descendant of the task's cgroup */
366 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
369 * Control Group subsystem type.
370 * See Documentation/cgroups/cgroups.txt for details
373 struct cgroup_subsys {
374 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
375 struct cgroup *cgrp);
376 int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
377 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
378 int (*can_attach)(struct cgroup_subsys *ss,
379 struct cgroup *cgrp, struct task_struct *tsk);
380 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
381 struct cgroup *old_cgrp, struct task_struct *tsk);
382 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
383 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
384 int (*populate)(struct cgroup_subsys *ss,
385 struct cgroup *cgrp);
386 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
387 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
394 * True if this subsys uses ID. ID is not available before cgroup_init()
395 * (not available in early_init time.)
398 #define MAX_CGROUP_TYPE_NAMELEN 32
402 * Protects sibling/children links of cgroups in this
403 * hierarchy, plus protects which hierarchy (or none) the
404 * subsystem is a part of (i.e. root/sibling). To avoid
405 * potential deadlocks, the following operations should not be
406 * undertaken while holding any hierarchy_mutex:
408 * - allocating memory
409 * - initiating hotplug events
411 struct mutex hierarchy_mutex;
412 struct lock_class_key subsys_key;
415 * Link to parent, and list entry in parent's children.
416 * Protected by this->hierarchy_mutex and cgroup_lock()
418 struct cgroupfs_root *root;
419 struct list_head sibling;
420 /* used when use_id == true */
425 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
426 #include <linux/cgroup_subsys.h>
429 static inline struct cgroup_subsys_state *cgroup_subsys_state(
430 struct cgroup *cgrp, int subsys_id)
432 return cgrp->subsys[subsys_id];
435 static inline struct cgroup_subsys_state *task_subsys_state(
436 struct task_struct *task, int subsys_id)
438 return rcu_dereference(task->cgroups->subsys[subsys_id]);
441 static inline struct cgroup* task_cgroup(struct task_struct *task,
444 return task_subsys_state(task, subsys_id)->cgroup;
447 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
450 /* A cgroup_iter should be treated as an opaque object */
452 struct list_head *cg_link;
453 struct list_head *task;
457 * To iterate across the tasks in a cgroup:
459 * 1) call cgroup_iter_start to intialize an iterator
461 * 2) call cgroup_iter_next() to retrieve member tasks until it
462 * returns NULL or until you want to end the iteration
464 * 3) call cgroup_iter_end() to destroy the iterator.
466 * Or, call cgroup_scan_tasks() to iterate through every task in a
467 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
468 * the test_task() callback, but not while calling the process_task()
471 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
472 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
473 struct cgroup_iter *it);
474 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
475 int cgroup_scan_tasks(struct cgroup_scanner *scan);
476 int cgroup_attach_task(struct cgroup *, struct task_struct *);
479 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
480 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
481 * CSS ID is assigned at cgroup allocation (create) automatically
482 * and removed when subsys calls free_css_id() function. This is because
483 * the lifetime of cgroup_subsys_state is subsys's matter.
485 * Looking up and scanning function should be called under rcu_read_lock().
486 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
487 * But the css returned by this routine can be "not populated yet" or "being
488 * destroyed". The caller should check css and cgroup's status.
492 * Typically Called at ->destroy(), or somewhere the subsys frees
493 * cgroup_subsys_state.
495 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
497 /* Find a cgroup_subsys_state which has given ID */
499 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
502 * Get a cgroup whose id is greater than or equal to id under tree of root.
503 * Returning a cgroup_subsys_state or NULL.
505 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
506 struct cgroup_subsys_state *root, int *foundid);
508 /* Returns true if root is ancestor of cg */
509 bool css_is_ancestor(struct cgroup_subsys_state *cg,
510 const struct cgroup_subsys_state *root);
512 /* Get id and depth of css */
513 unsigned short css_id(struct cgroup_subsys_state *css);
514 unsigned short css_depth(struct cgroup_subsys_state *css);
516 #else /* !CONFIG_CGROUPS */
518 static inline int cgroup_init_early(void) { return 0; }
519 static inline int cgroup_init(void) { return 0; }
520 static inline void cgroup_fork(struct task_struct *p) {}
521 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
522 static inline void cgroup_post_fork(struct task_struct *p) {}
523 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
525 static inline void cgroup_lock(void) {}
526 static inline void cgroup_unlock(void) {}
527 static inline int cgroupstats_build(struct cgroupstats *stats,
528 struct dentry *dentry)
533 #endif /* !CONFIG_CGROUPS */
535 #endif /* _LINUX_CGROUP_H */