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>
20 #include <linux/xattr.h>
30 extern int cgroup_init_early(void);
31 extern int cgroup_init(void);
32 extern void cgroup_lock(void);
33 extern int cgroup_lock_is_held(void);
34 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
35 extern void cgroup_unlock(void);
36 extern void cgroup_fork(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 #define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
49 enum cgroup_subsys_id {
50 #include <linux/cgroup_subsys.h>
53 #undef IS_SUBSYS_ENABLED
56 /* Per-subsystem/per-cgroup state maintained by the system. */
57 struct cgroup_subsys_state {
59 * The cgroup that this subsystem is attached to. Useful
60 * for subsystems that want to know about the cgroup
63 struct cgroup *cgroup;
66 * State maintained by the cgroup system to allow subsystems
67 * to be "busy". Should be accessed via css_get(),
68 * css_tryget() and and css_put().
74 /* ID for this css, if possible */
75 struct css_id __rcu *id;
77 /* Used to put @cgroup->dentry on the last css_put() */
78 struct work_struct dput_work;
81 /* bits in struct cgroup_subsys_state flags field */
83 CSS_ROOT, /* This CSS is the root of the subsystem */
86 /* Caller must verify that the css is not for root cgroup */
87 static inline void __css_get(struct cgroup_subsys_state *css, int count)
89 atomic_add(count, &css->refcnt);
93 * Call css_get() to hold a reference on the css; it can be used
94 * for a reference obtained via:
95 * - an existing ref-counted reference to the css
96 * - task->cgroups for a locked task
99 static inline void css_get(struct cgroup_subsys_state *css)
101 /* We don't need to reference count the root state */
102 if (!test_bit(CSS_ROOT, &css->flags))
107 * Call css_tryget() to take a reference on a css if your existing
108 * (known-valid) reference isn't already ref-counted. Returns false if
109 * the css has been destroyed.
112 extern bool __css_tryget(struct cgroup_subsys_state *css);
113 static inline bool css_tryget(struct cgroup_subsys_state *css)
115 if (test_bit(CSS_ROOT, &css->flags))
117 return __css_tryget(css);
121 * css_put() should be called to release a reference taken by
122 * css_get() or css_tryget()
125 extern void __css_put(struct cgroup_subsys_state *css);
126 static inline void css_put(struct cgroup_subsys_state *css)
128 if (!test_bit(CSS_ROOT, &css->flags))
132 /* bits in struct cgroup flags field */
134 /* Control Group is dead */
137 * Control Group has previously had a child cgroup or a task,
138 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
141 /* Control Group requires release notifications to userspace */
142 CGRP_NOTIFY_ON_RELEASE,
144 * Clone cgroup values when creating a new child cgroup
150 unsigned long flags; /* "unsigned long" so bitops work */
153 * count users of this cgroup. >0 means busy, but doesn't
154 * necessarily indicate the number of tasks in the cgroup
159 * We link our 'sibling' struct into our parent's 'children'.
160 * Our children link their 'sibling' into our 'children'.
162 struct list_head sibling; /* my parent's children */
163 struct list_head children; /* my children */
164 struct list_head files; /* my files */
166 struct cgroup *parent; /* my parent */
167 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
169 /* Private pointers for each registered subsystem */
170 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
172 struct cgroupfs_root *root;
173 struct cgroup *top_cgroup;
176 * List of cg_cgroup_links pointing at css_sets with
177 * tasks in this cgroup. Protected by css_set_lock
179 struct list_head css_sets;
181 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
182 struct list_head cft_q_node; /* used during cftype add/rm */
185 * Linked list running through all cgroups that can
186 * potentially be reaped by the release agent. Protected by
189 struct list_head release_list;
192 * list of pidlists, up to two for each namespace (one for procs, one
193 * for tasks); created on demand.
195 struct list_head pidlists;
196 struct mutex pidlist_mutex;
198 /* For RCU-protected deletion */
199 struct rcu_head rcu_head;
201 /* List of events which userspace want to receive */
202 struct list_head event_list;
203 spinlock_t event_list_lock;
205 /* directory xattrs */
206 struct simple_xattrs xattrs;
210 * A css_set is a structure holding pointers to a set of
211 * cgroup_subsys_state objects. This saves space in the task struct
212 * object and speeds up fork()/exit(), since a single inc/dec and a
213 * list_add()/del() can bump the reference count on the entire cgroup
219 /* Reference count */
223 * List running through all cgroup groups in the same hash
224 * slot. Protected by css_set_lock
226 struct hlist_node hlist;
229 * List running through all tasks using this cgroup
230 * group. Protected by css_set_lock
232 struct list_head tasks;
235 * List of cg_cgroup_link objects on link chains from
236 * cgroups referenced from this css_set. Protected by
239 struct list_head cg_links;
242 * Set of subsystem states, one for each subsystem. This array
243 * is immutable after creation apart from the init_css_set
244 * during subsystem registration (at boot time) and modular subsystem
247 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
249 /* For RCU-protected deletion */
250 struct rcu_head rcu_head;
254 * cgroup_map_cb is an abstract callback API for reporting map-valued
258 struct cgroup_map_cb {
259 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
264 * struct cftype: handler definitions for cgroup control files
266 * When reading/writing to a file:
267 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
268 * - the 'cftype' of the file is file->f_dentry->d_fsdata
272 #define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
273 #define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
275 #define MAX_CFTYPE_NAME 64
279 * By convention, the name should begin with the name of the
280 * subsystem, followed by a period. Zero length string indicates
281 * end of cftype array.
283 char name[MAX_CFTYPE_NAME];
286 * If not 0, file mode is set to this value, otherwise it will
287 * be figured out automatically
292 * If non-zero, defines the maximum length of string that can
293 * be passed to write_string; defaults to 64
295 size_t max_write_len;
301 struct simple_xattrs xattrs;
303 int (*open)(struct inode *inode, struct file *file);
304 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
306 char __user *buf, size_t nbytes, loff_t *ppos);
308 * read_u64() is a shortcut for the common case of returning a
309 * single integer. Use it in place of read()
311 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
313 * read_s64() is a signed version of read_u64()
315 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
317 * read_map() is used for defining a map of key/value
318 * pairs. It should call cb->fill(cb, key, value) for each
319 * entry. The key/value pairs (and their ordering) should not
320 * change between reboots.
322 int (*read_map)(struct cgroup *cont, struct cftype *cft,
323 struct cgroup_map_cb *cb);
325 * read_seq_string() is used for outputting a simple sequence
328 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
331 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
333 const char __user *buf, size_t nbytes, loff_t *ppos);
336 * write_u64() is a shortcut for the common case of accepting
337 * a single integer (as parsed by simple_strtoull) from
338 * userspace. Use in place of write(); return 0 or error.
340 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
342 * write_s64() is a signed version of write_u64()
344 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
347 * write_string() is passed a nul-terminated kernelspace
348 * buffer of maximum length determined by max_write_len.
349 * Returns 0 or -ve error code.
351 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
354 * trigger() callback can be used to get some kick from the
355 * userspace, when the actual string written is not important
356 * at all. The private field can be used to determine the
357 * kick type for multiplexing.
359 int (*trigger)(struct cgroup *cgrp, unsigned int event);
361 int (*release)(struct inode *inode, struct file *file);
364 * register_event() callback will be used to add new userspace
365 * waiter for changes related to the cftype. Implement it if
366 * you want to provide this functionality. Use eventfd_signal()
367 * on eventfd to send notification to userspace.
369 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
370 struct eventfd_ctx *eventfd, const char *args);
372 * unregister_event() callback will be called when userspace
373 * closes the eventfd or on cgroup removing.
374 * This callback must be implemented, if you want provide
375 * notification functionality.
377 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
378 struct eventfd_ctx *eventfd);
382 * cftype_sets describe cftypes belonging to a subsystem and are chained at
383 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
384 * terminated by zero length name.
387 struct list_head node; /* chained at subsys->cftsets */
391 struct cgroup_scanner {
393 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
394 void (*process_task)(struct task_struct *p,
395 struct cgroup_scanner *scan);
396 struct ptr_heap *heap;
400 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
401 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
403 int cgroup_is_removed(const struct cgroup *cgrp);
405 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
407 int cgroup_task_count(const struct cgroup *cgrp);
409 /* Return true if cgrp is a descendant of the task's cgroup */
410 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
413 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
416 struct cgroup_taskset;
417 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
418 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
419 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
420 int cgroup_taskset_size(struct cgroup_taskset *tset);
423 * cgroup_taskset_for_each - iterate cgroup_taskset
424 * @task: the loop cursor
425 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
426 * @tset: taskset to iterate
428 #define cgroup_taskset_for_each(task, skip_cgrp, tset) \
429 for ((task) = cgroup_taskset_first((tset)); (task); \
430 (task) = cgroup_taskset_next((tset))) \
431 if (!(skip_cgrp) || \
432 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
435 * Control Group subsystem type.
436 * See Documentation/cgroups/cgroups.txt for details
439 struct cgroup_subsys {
440 struct cgroup_subsys_state *(*create)(struct cgroup *cgrp);
441 void (*pre_destroy)(struct cgroup *cgrp);
442 void (*destroy)(struct cgroup *cgrp);
443 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
444 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
445 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
446 void (*fork)(struct task_struct *task);
447 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
448 struct task_struct *task);
449 void (*post_clone)(struct cgroup *cgrp);
450 void (*bind)(struct cgroup *root);
457 * True if this subsys uses ID. ID is not available before cgroup_init()
458 * (not available in early_init time.)
463 * If %false, this subsystem is properly hierarchical -
464 * configuration, resource accounting and restriction on a parent
465 * cgroup cover those of its children. If %true, hierarchy support
466 * is broken in some ways - some subsystems ignore hierarchy
467 * completely while others are only implemented half-way.
469 * It's now disallowed to create nested cgroups if the subsystem is
470 * broken and cgroup core will emit a warning message on such
471 * cases. Eventually, all subsystems will be made properly
472 * hierarchical and this will go away.
474 bool broken_hierarchy;
475 bool warned_broken_hierarchy;
477 #define MAX_CGROUP_TYPE_NAMELEN 32
481 * Link to parent, and list entry in parent's children.
482 * Protected by cgroup_lock()
484 struct cgroupfs_root *root;
485 struct list_head sibling;
486 /* used when use_id == true */
490 /* list of cftype_sets */
491 struct list_head cftsets;
493 /* base cftypes, automatically [de]registered with subsys itself */
494 struct cftype *base_cftypes;
495 struct cftype_set base_cftset;
497 /* should be defined only by modular subsystems */
498 struct module *module;
501 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
502 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
503 #include <linux/cgroup_subsys.h>
504 #undef IS_SUBSYS_ENABLED
507 static inline struct cgroup_subsys_state *cgroup_subsys_state(
508 struct cgroup *cgrp, int subsys_id)
510 return cgrp->subsys[subsys_id];
514 * function to get the cgroup_subsys_state which allows for extra
515 * rcu_dereference_check() conditions, such as locks used during the
516 * cgroup_subsys::attach() methods.
518 #define task_subsys_state_check(task, subsys_id, __c) \
519 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
520 lockdep_is_held(&task->alloc_lock) || \
521 cgroup_lock_is_held() || (__c))
523 static inline struct cgroup_subsys_state *
524 task_subsys_state(struct task_struct *task, int subsys_id)
526 return task_subsys_state_check(task, subsys_id, false);
529 static inline struct cgroup* task_cgroup(struct task_struct *task,
532 return task_subsys_state(task, subsys_id)->cgroup;
535 /* A cgroup_iter should be treated as an opaque object */
537 struct list_head *cg_link;
538 struct list_head *task;
542 * To iterate across the tasks in a cgroup:
544 * 1) call cgroup_iter_start to initialize an iterator
546 * 2) call cgroup_iter_next() to retrieve member tasks until it
547 * returns NULL or until you want to end the iteration
549 * 3) call cgroup_iter_end() to destroy the iterator.
551 * Or, call cgroup_scan_tasks() to iterate through every task in a
552 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
553 * the test_task() callback, but not while calling the process_task()
556 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
557 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
558 struct cgroup_iter *it);
559 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
560 int cgroup_scan_tasks(struct cgroup_scanner *scan);
561 int cgroup_attach_task(struct cgroup *, struct task_struct *);
562 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
565 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
566 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
567 * CSS ID is assigned at cgroup allocation (create) automatically
568 * and removed when subsys calls free_css_id() function. This is because
569 * the lifetime of cgroup_subsys_state is subsys's matter.
571 * Looking up and scanning function should be called under rcu_read_lock().
572 * Taking cgroup_mutex is not necessary for following calls.
573 * But the css returned by this routine can be "not populated yet" or "being
574 * destroyed". The caller should check css and cgroup's status.
578 * Typically Called at ->destroy(), or somewhere the subsys frees
579 * cgroup_subsys_state.
581 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
583 /* Find a cgroup_subsys_state which has given ID */
585 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
588 * Get a cgroup whose id is greater than or equal to id under tree of root.
589 * Returning a cgroup_subsys_state or NULL.
591 struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
592 struct cgroup_subsys_state *root, int *foundid);
594 /* Returns true if root is ancestor of cg */
595 bool css_is_ancestor(struct cgroup_subsys_state *cg,
596 const struct cgroup_subsys_state *root);
598 /* Get id and depth of css */
599 unsigned short css_id(struct cgroup_subsys_state *css);
600 unsigned short css_depth(struct cgroup_subsys_state *css);
601 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
603 #else /* !CONFIG_CGROUPS */
605 static inline int cgroup_init_early(void) { return 0; }
606 static inline int cgroup_init(void) { return 0; }
607 static inline void cgroup_fork(struct task_struct *p) {}
608 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
609 static inline void cgroup_post_fork(struct task_struct *p) {}
610 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
612 static inline void cgroup_lock(void) {}
613 static inline void cgroup_unlock(void) {}
614 static inline int cgroupstats_build(struct cgroupstats *stats,
615 struct dentry *dentry)
620 /* No cgroups - nothing to do */
621 static inline int cgroup_attach_task_all(struct task_struct *from,
622 struct task_struct *t)
627 #endif /* !CONFIG_CGROUPS */
629 #endif /* _LINUX_CGROUP_H */