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 * Clone cgroup values when creating a new child cgroup
154 unsigned long flags; /* "unsigned long" so bitops work */
157 * count users of this cgroup. >0 means busy, but doesn't
158 * necessarily indicate the number of tasks in the cgroup
163 * We link our 'sibling' struct into our parent's 'children'.
164 * Our children link their 'sibling' into our 'children'.
166 struct list_head sibling; /* my parent's children */
167 struct list_head children; /* my children */
168 struct list_head files; /* my files */
170 struct cgroup *parent; /* my parent */
171 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
173 /* Private pointers for each registered subsystem */
174 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
176 struct cgroupfs_root *root;
177 struct cgroup *top_cgroup;
180 * List of cg_cgroup_links pointing at css_sets with
181 * tasks in this cgroup. Protected by css_set_lock
183 struct list_head css_sets;
185 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
186 struct list_head cft_q_node; /* used during cftype add/rm */
189 * Linked list running through all cgroups that can
190 * potentially be reaped by the release agent. Protected by
193 struct list_head release_list;
196 * list of pidlists, up to two for each namespace (one for procs, one
197 * for tasks); created on demand.
199 struct list_head pidlists;
200 struct mutex pidlist_mutex;
202 /* For RCU-protected deletion */
203 struct rcu_head rcu_head;
205 /* List of events which userspace want to receive */
206 struct list_head event_list;
207 spinlock_t event_list_lock;
211 * A css_set is a structure holding pointers to a set of
212 * cgroup_subsys_state objects. This saves space in the task struct
213 * object and speeds up fork()/exit(), since a single inc/dec and a
214 * list_add()/del() can bump the reference count on the entire cgroup
220 /* Reference count */
224 * List running through all cgroup groups in the same hash
225 * slot. Protected by css_set_lock
227 struct hlist_node hlist;
230 * List running through all tasks using this cgroup
231 * group. Protected by css_set_lock
233 struct list_head tasks;
236 * List of cg_cgroup_link objects on link chains from
237 * cgroups referenced from this css_set. Protected by
240 struct list_head cg_links;
243 * Set of subsystem states, one for each subsystem. This array
244 * is immutable after creation apart from the init_css_set
245 * during subsystem registration (at boot time) and modular subsystem
248 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
250 /* For RCU-protected deletion */
251 struct rcu_head rcu_head;
255 * cgroup_map_cb is an abstract callback API for reporting map-valued
259 struct cgroup_map_cb {
260 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
265 * struct cftype: handler definitions for cgroup control files
267 * When reading/writing to a file:
268 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
269 * - the 'cftype' of the file is file->f_dentry->d_fsdata
273 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
274 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
276 #define MAX_CFTYPE_NAME 64
280 * By convention, the name should begin with the name of the
281 * subsystem, followed by a period. Zero length string indicates
282 * end of cftype array.
284 char name[MAX_CFTYPE_NAME];
287 * If not 0, file mode is set to this value, otherwise it will
288 * be figured out automatically
293 * If non-zero, defines the maximum length of string that can
294 * be passed to write_string; defaults to 64
296 size_t max_write_len;
301 int (*open)(struct inode *inode, struct file *file);
302 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
304 char __user *buf, size_t nbytes, loff_t *ppos);
306 * read_u64() is a shortcut for the common case of returning a
307 * single integer. Use it in place of read()
309 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
311 * read_s64() is a signed version of read_u64()
313 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
315 * read_map() is used for defining a map of key/value
316 * pairs. It should call cb->fill(cb, key, value) for each
317 * entry. The key/value pairs (and their ordering) should not
318 * change between reboots.
320 int (*read_map)(struct cgroup *cont, struct cftype *cft,
321 struct cgroup_map_cb *cb);
323 * read_seq_string() is used for outputting a simple sequence
326 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
329 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
331 const char __user *buf, size_t nbytes, loff_t *ppos);
334 * write_u64() is a shortcut for the common case of accepting
335 * a single integer (as parsed by simple_strtoull) from
336 * userspace. Use in place of write(); return 0 or error.
338 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
340 * write_s64() is a signed version of write_u64()
342 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
345 * write_string() is passed a nul-terminated kernelspace
346 * buffer of maximum length determined by max_write_len.
347 * Returns 0 or -ve error code.
349 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
352 * trigger() callback can be used to get some kick from the
353 * userspace, when the actual string written is not important
354 * at all. The private field can be used to determine the
355 * kick type for multiplexing.
357 int (*trigger)(struct cgroup *cgrp, unsigned int event);
359 int (*release)(struct inode *inode, struct file *file);
362 * register_event() callback will be used to add new userspace
363 * waiter for changes related to the cftype. Implement it if
364 * you want to provide this functionality. Use eventfd_signal()
365 * on eventfd to send notification to userspace.
367 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
368 struct eventfd_ctx *eventfd, const char *args);
370 * unregister_event() callback will be called when userspace
371 * closes the eventfd or on cgroup removing.
372 * This callback must be implemented, if you want provide
373 * notification functionality.
375 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
376 struct eventfd_ctx *eventfd);
380 * cftype_sets describe cftypes belonging to a subsystem and are chained at
381 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
382 * terminated by zero length name.
385 struct list_head node; /* chained at subsys->cftsets */
386 const struct cftype *cfts;
389 struct cgroup_scanner {
391 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
392 void (*process_task)(struct task_struct *p,
393 struct cgroup_scanner *scan);
394 struct ptr_heap *heap;
398 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts);
399 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts);
401 int cgroup_is_removed(const struct cgroup *cgrp);
403 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
405 int cgroup_task_count(const struct cgroup *cgrp);
407 /* Return true if cgrp is a descendant of the task's cgroup */
408 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
411 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
414 struct cgroup_taskset;
415 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
416 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
417 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
418 int cgroup_taskset_size(struct cgroup_taskset *tset);
421 * cgroup_taskset_for_each - iterate cgroup_taskset
422 * @task: the loop cursor
423 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
424 * @tset: taskset to iterate
426 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
427 for ((task) = cgroup_taskset_first((tset)); (task); \
428 (task) = cgroup_taskset_next((tset))) \
429 if (!(skip_cgrp) || \
430 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
433 * Control Group subsystem type.
434 * See Documentation/cgroups/cgroups.txt for details
437 struct cgroup_subsys {
438 struct cgroup_subsys_state *(*create)(struct cgroup *cgrp);
439 int (*pre_destroy)(struct cgroup *cgrp);
440 void (*destroy)(struct cgroup *cgrp);
441 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
442 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
443 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
444 void (*fork)(struct task_struct *task);
445 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
446 struct task_struct *task);
447 void (*post_clone)(struct cgroup *cgrp);
448 void (*bind)(struct cgroup *root);
455 * True if this subsys uses ID. ID is not available before cgroup_init()
456 * (not available in early_init time.)
460 #define MAX_CGROUP_TYPE_NAMELEN 32
464 * Link to parent, and list entry in parent's children.
465 * Protected by cgroup_lock()
467 struct cgroupfs_root *root;
468 struct list_head sibling;
469 /* used when use_id == true */
473 /* list of cftype_sets */
474 struct list_head cftsets;
476 /* base cftypes, automatically [de]registered with subsys itself */
477 struct cftype *base_cftypes;
478 struct cftype_set base_cftset;
480 /* should be defined only by modular subsystems */
481 struct module *module;
484 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
485 #include <linux/cgroup_subsys.h>
488 static inline struct cgroup_subsys_state *cgroup_subsys_state(
489 struct cgroup *cgrp, int subsys_id)
491 return cgrp->subsys[subsys_id];
495 * function to get the cgroup_subsys_state which allows for extra
496 * rcu_dereference_check() conditions, such as locks used during the
497 * cgroup_subsys::attach() methods.
499 #define task_subsys_state_check(task, subsys_id, __c) \
500 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
501 lockdep_is_held(&task->alloc_lock) || \
502 cgroup_lock_is_held() || (__c))
504 static inline struct cgroup_subsys_state *
505 task_subsys_state(struct task_struct *task, int subsys_id)
507 return task_subsys_state_check(task, subsys_id, false);
510 static inline struct cgroup* task_cgroup(struct task_struct *task,
513 return task_subsys_state(task, subsys_id)->cgroup;
516 /* A cgroup_iter should be treated as an opaque object */
518 struct list_head *cg_link;
519 struct list_head *task;
523 * To iterate across the tasks in a cgroup:
525 * 1) call cgroup_iter_start to initialize an iterator
527 * 2) call cgroup_iter_next() to retrieve member tasks until it
528 * returns NULL or until you want to end the iteration
530 * 3) call cgroup_iter_end() to destroy the iterator.
532 * Or, call cgroup_scan_tasks() to iterate through every task in a
533 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
534 * the test_task() callback, but not while calling the process_task()
537 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
538 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
539 struct cgroup_iter *it);
540 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
541 int cgroup_scan_tasks(struct cgroup_scanner *scan);
542 int cgroup_attach_task(struct cgroup *, struct task_struct *);
543 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
546 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
547 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
548 * CSS ID is assigned at cgroup allocation (create) automatically
549 * and removed when subsys calls free_css_id() function. This is because
550 * the lifetime of cgroup_subsys_state is subsys's matter.
552 * Looking up and scanning function should be called under rcu_read_lock().
553 * Taking cgroup_mutex is not necessary for following calls.
554 * But the css returned by this routine can be "not populated yet" or "being
555 * destroyed". The caller should check css and cgroup's status.
559 * Typically Called at ->destroy(), or somewhere the subsys frees
560 * cgroup_subsys_state.
562 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
564 /* Find a cgroup_subsys_state which has given ID */
566 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
569 * Get a cgroup whose id is greater than or equal to id under tree of root.
570 * Returning a cgroup_subsys_state or NULL.
572 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
573 struct cgroup_subsys_state *root, int *foundid);
575 /* Returns true if root is ancestor of cg */
576 bool css_is_ancestor(struct cgroup_subsys_state *cg,
577 const struct cgroup_subsys_state *root);
579 /* Get id and depth of css */
580 unsigned short css_id(struct cgroup_subsys_state *css);
581 unsigned short css_depth(struct cgroup_subsys_state *css);
582 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
584 #else /* !CONFIG_CGROUPS */
586 static inline int cgroup_init_early(void) { return 0; }
587 static inline int cgroup_init(void) { return 0; }
588 static inline void cgroup_fork(struct task_struct *p) {}
589 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
590 static inline void cgroup_post_fork(struct task_struct *p) {}
591 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
593 static inline void cgroup_lock(void) {}
594 static inline void cgroup_unlock(void) {}
595 static inline int cgroupstats_build(struct cgroupstats *stats,
596 struct dentry *dentry)
601 /* No cgroups - nothing to do */
602 static inline int cgroup_attach_task_all(struct task_struct *from,
603 struct task_struct *t)
608 #endif /* !CONFIG_CGROUPS */
610 #endif /* _LINUX_CGROUP_H */