blob: d2f82979f6c1bc68316e894285275d1d422008e4 [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001#ifndef _LINUX_CGROUP_H
2#define _LINUX_CGROUP_H
3/*
4 * cgroup interface
5 *
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
8 *
9 */
10
11#include <linux/sched.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070012#include <linux/cpumask.h>
13#include <linux/nodemask.h>
14#include <linux/rcupdate.h>
Tejun Heoeb6fd502012-11-09 09:12:29 -080015#include <linux/rculist.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070016#include <linux/cgroupstats.h>
Cliff Wickman31a7df02008-02-07 00:14:42 -080017#include <linux/prio_heap.h>
Paul Menagecc31edc2008-10-18 20:28:04 -070018#include <linux/rwsem.h>
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070019#include <linux/idr.h>
Tejun Heo48ddbe12012-04-01 12:09:56 -070020#include <linux/workqueue.h>
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -040021#include <linux/xattr.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070022
23#ifdef CONFIG_CGROUPS
24
25struct cgroupfs_root;
26struct cgroup_subsys;
27struct inode;
Paul Menage84eea842008-07-25 01:47:00 -070028struct cgroup;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070029struct css_id;
Paul Menageddbcc7e2007-10-18 23:39:30 -070030
31extern int cgroup_init_early(void);
32extern int cgroup_init(void);
Paul Menageddbcc7e2007-10-18 23:39:30 -070033extern void cgroup_lock(void);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -080034extern int cgroup_lock_is_held(void);
Paul Menage84eea842008-07-25 01:47:00 -070035extern bool cgroup_lock_live_group(struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -070036extern void cgroup_unlock(void);
Paul Menageb4f48b62007-10-18 23:39:33 -070037extern void cgroup_fork(struct task_struct *p);
Paul Menage817929e2007-10-18 23:39:36 -070038extern void cgroup_post_fork(struct task_struct *p);
Paul Menageb4f48b62007-10-18 23:39:33 -070039extern void cgroup_exit(struct task_struct *p, int run_callbacks);
Balbir Singh846c7bb2007-10-18 23:39:44 -070040extern int cgroupstats_build(struct cgroupstats *stats,
41 struct dentry *dentry);
Ben Blume6a11052010-03-10 15:22:09 -080042extern int cgroup_load_subsys(struct cgroup_subsys *ss);
Ben Blumcf5d5942010-03-10 15:22:09 -080043extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
Paul Menageddbcc7e2007-10-18 23:39:30 -070044
Alexey Dobriyan828c0952009-10-01 15:43:56 -070045extern const struct file_operations proc_cgroup_operations;
Paul Menagea4243162007-10-18 23:39:35 -070046
Ben Blumaae8aab2010-03-10 15:22:07 -080047/* Define the enumeration of all builtin cgroup subsystems */
Paul Menage817929e2007-10-18 23:39:36 -070048#define SUBSYS(_x) _x ## _subsys_id,
Daniel Wagner8a8e04d2012-09-12 16:12:07 +020049#define IS_SUBSYS_ENABLED(option) IS_ENABLED(option)
Paul Menage817929e2007-10-18 23:39:36 -070050enum cgroup_subsys_id {
51#include <linux/cgroup_subsys.h>
Daniel Wagnera6f00292012-09-12 16:12:08 +020052 CGROUP_SUBSYS_COUNT,
Paul Menage817929e2007-10-18 23:39:36 -070053};
Daniel Wagner5fc0b022012-09-12 16:12:05 +020054#undef IS_SUBSYS_ENABLED
Paul Menage817929e2007-10-18 23:39:36 -070055#undef SUBSYS
56
Paul Menageddbcc7e2007-10-18 23:39:30 -070057/* Per-subsystem/per-cgroup state maintained by the system. */
58struct cgroup_subsys_state {
Paul Menaged20a3902009-04-02 16:57:22 -070059 /*
60 * The cgroup that this subsystem is attached to. Useful
Paul Menageddbcc7e2007-10-18 23:39:30 -070061 * for subsystems that want to know about the cgroup
Paul Menaged20a3902009-04-02 16:57:22 -070062 * hierarchy structure
63 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070064 struct cgroup *cgroup;
65
Paul Menaged20a3902009-04-02 16:57:22 -070066 /*
67 * State maintained by the cgroup system to allow subsystems
Paul Menagee7c5ec92009-01-07 18:08:38 -080068 * to be "busy". Should be accessed via css_get(),
Paul Menaged20a3902009-04-02 16:57:22 -070069 * css_tryget() and and css_put().
70 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070071
72 atomic_t refcnt;
73
74 unsigned long flags;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070075 /* ID for this css, if possible */
Arnd Bergmann2c392b82010-02-24 19:41:39 +010076 struct css_id __rcu *id;
Tejun Heo48ddbe12012-04-01 12:09:56 -070077
78 /* Used to put @cgroup->dentry on the last css_put() */
79 struct work_struct dput_work;
Paul Menageddbcc7e2007-10-18 23:39:30 -070080};
81
82/* bits in struct cgroup_subsys_state flags field */
83enum {
Tejun Heo38b53ab2012-11-19 08:13:36 -080084 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
Tejun Heo92fb9742012-11-19 08:13:38 -080085 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
Paul Menageddbcc7e2007-10-18 23:39:30 -070086};
87
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -080088/* Caller must verify that the css is not for root cgroup */
89static inline void __css_get(struct cgroup_subsys_state *css, int count)
90{
91 atomic_add(count, &css->refcnt);
92}
93
Paul Menageddbcc7e2007-10-18 23:39:30 -070094/*
Paul Menagee7c5ec92009-01-07 18:08:38 -080095 * Call css_get() to hold a reference on the css; it can be used
96 * for a reference obtained via:
97 * - an existing ref-counted reference to the css
98 * - task->cgroups for a locked task
Paul Menageddbcc7e2007-10-18 23:39:30 -070099 */
100
101static inline void css_get(struct cgroup_subsys_state *css)
102{
103 /* We don't need to reference count the root state */
Tejun Heo38b53ab2012-11-19 08:13:36 -0800104 if (!(css->flags & CSS_ROOT))
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -0800105 __css_get(css, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700106}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800107
Paul Menagee7c5ec92009-01-07 18:08:38 -0800108/*
109 * Call css_tryget() to take a reference on a css if your existing
110 * (known-valid) reference isn't already ref-counted. Returns false if
111 * the css has been destroyed.
112 */
113
Tejun Heo28b4c272012-04-01 12:09:56 -0700114extern bool __css_tryget(struct cgroup_subsys_state *css);
Paul Menagee7c5ec92009-01-07 18:08:38 -0800115static inline bool css_tryget(struct cgroup_subsys_state *css)
116{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800117 if (css->flags & CSS_ROOT)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800118 return true;
Tejun Heo28b4c272012-04-01 12:09:56 -0700119 return __css_tryget(css);
Paul Menagee7c5ec92009-01-07 18:08:38 -0800120}
121
Paul Menageddbcc7e2007-10-18 23:39:30 -0700122/*
123 * css_put() should be called to release a reference taken by
Paul Menagee7c5ec92009-01-07 18:08:38 -0800124 * css_get() or css_tryget()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700125 */
126
Tejun Heo28b4c272012-04-01 12:09:56 -0700127extern void __css_put(struct cgroup_subsys_state *css);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700128static inline void css_put(struct cgroup_subsys_state *css)
129{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800130 if (!(css->flags & CSS_ROOT))
Tejun Heo28b4c272012-04-01 12:09:56 -0700131 __css_put(css);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700132}
133
Paul Menage3116f0e2008-04-29 01:00:04 -0700134/* bits in struct cgroup flags field */
135enum {
136 /* Control Group is dead */
137 CGRP_REMOVED,
Paul Menaged20a3902009-04-02 16:57:22 -0700138 /*
139 * Control Group has previously had a child cgroup or a task,
140 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
141 */
Paul Menage3116f0e2008-04-29 01:00:04 -0700142 CGRP_RELEASABLE,
143 /* Control Group requires release notifications to userspace */
144 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700145 /*
Tejun Heo2260e7f2012-11-19 08:13:38 -0800146 * Clone the parent's configuration when creating a new child
147 * cpuset cgroup. For historical reasons, this option can be
148 * specified at mount time and thus is implemented here.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700149 */
Tejun Heo2260e7f2012-11-19 08:13:38 -0800150 CGRP_CPUSET_CLONE_CHILDREN,
Paul Menage3116f0e2008-04-29 01:00:04 -0700151};
152
Paul Menageddbcc7e2007-10-18 23:39:30 -0700153struct cgroup {
154 unsigned long flags; /* "unsigned long" so bitops work */
155
Paul Menaged20a3902009-04-02 16:57:22 -0700156 /*
157 * count users of this cgroup. >0 means busy, but doesn't
158 * necessarily indicate the number of tasks in the cgroup
159 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700160 atomic_t count;
161
162 /*
163 * We link our 'sibling' struct into our parent's 'children'.
164 * Our children link their 'sibling' into our 'children'.
165 */
166 struct list_head sibling; /* my parent's children */
167 struct list_head children; /* my children */
Tejun Heo05ef1d72012-04-01 12:09:56 -0700168 struct list_head files; /* my files */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700169
Paul Menaged20a3902009-04-02 16:57:22 -0700170 struct cgroup *parent; /* my parent */
Tejun Heofebfcef2012-11-19 08:13:36 -0800171 struct dentry *dentry; /* cgroup fs entry, RCU protected */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700172
173 /* Private pointers for each registered subsystem */
174 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
175
176 struct cgroupfs_root *root;
177 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700178
179 /*
180 * List of cg_cgroup_links pointing at css_sets with
181 * tasks in this cgroup. Protected by css_set_lock
182 */
183 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700184
Tejun Heob0ca5a82012-04-01 12:09:54 -0700185 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700186 struct list_head cft_q_node; /* used during cftype add/rm */
Tejun Heob0ca5a82012-04-01 12:09:54 -0700187
Paul Menage81a6a5c2007-10-18 23:39:38 -0700188 /*
189 * Linked list running through all cgroups that can
190 * potentially be reaped by the release agent. Protected by
191 * release_list_lock
192 */
193 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700194
Ben Blum72a8cb32009-09-23 15:56:27 -0700195 /*
196 * list of pidlists, up to two for each namespace (one for procs, one
197 * for tasks); created on demand.
198 */
199 struct list_head pidlists;
200 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800201
202 /* For RCU-protected deletion */
203 struct rcu_head rcu_head;
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800204
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300205 /* List of events which userspace want to receive */
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800206 struct list_head event_list;
207 spinlock_t event_list_lock;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400208
209 /* directory xattrs */
210 struct simple_xattrs xattrs;
Paul Menage817929e2007-10-18 23:39:36 -0700211};
212
Paul Menaged20a3902009-04-02 16:57:22 -0700213/*
214 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700215 * cgroup_subsys_state objects. This saves space in the task struct
216 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700217 * list_add()/del() can bump the reference count on the entire cgroup
218 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700219 */
220
221struct css_set {
222
223 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700224 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700225
226 /*
Li Zefan472b1052008-04-29 01:00:11 -0700227 * List running through all cgroup groups in the same hash
228 * slot. Protected by css_set_lock
229 */
230 struct hlist_node hlist;
231
232 /*
Paul Menage817929e2007-10-18 23:39:36 -0700233 * List running through all tasks using this cgroup
234 * group. Protected by css_set_lock
235 */
236 struct list_head tasks;
237
238 /*
239 * List of cg_cgroup_link objects on link chains from
240 * cgroups referenced from this css_set. Protected by
241 * css_set_lock
242 */
243 struct list_head cg_links;
244
245 /*
246 * Set of subsystem states, one for each subsystem. This array
247 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800248 * during subsystem registration (at boot time) and modular subsystem
249 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700250 */
251 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700252
253 /* For RCU-protected deletion */
254 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700255};
256
Paul Menage91796562008-04-29 01:00:01 -0700257/*
258 * cgroup_map_cb is an abstract callback API for reporting map-valued
259 * control files
260 */
261
262struct cgroup_map_cb {
263 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
264 void *state;
265};
266
Paul Menaged20a3902009-04-02 16:57:22 -0700267/*
268 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700269 *
270 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800271 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700272 * - the 'cftype' of the file is file->f_dentry->d_fsdata
273 */
274
Tejun Heo8e3f6542012-04-01 12:09:55 -0700275/* cftype->flags */
276#define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
277#define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create onp root cg */
278
279#define MAX_CFTYPE_NAME 64
280
Paul Menageddbcc7e2007-10-18 23:39:30 -0700281struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700282 /*
283 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700284 * subsystem, followed by a period. Zero length string indicates
285 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700286 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700287 char name[MAX_CFTYPE_NAME];
288 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700289 /*
290 * If not 0, file mode is set to this value, otherwise it will
291 * be figured out automatically
292 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400293 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700294
295 /*
296 * If non-zero, defines the maximum length of string that can
297 * be passed to write_string; defaults to 64
298 */
299 size_t max_write_len;
300
Tejun Heo8e3f6542012-04-01 12:09:55 -0700301 /* CFTYPE_* flags */
302 unsigned int flags;
303
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400304 /* file xattrs */
305 struct simple_xattrs xattrs;
306
Paul Menagece16b492008-07-25 01:46:57 -0700307 int (*open)(struct inode *inode, struct file *file);
308 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
309 struct file *file,
310 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700311 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700312 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700313 * single integer. Use it in place of read()
314 */
Paul Menagece16b492008-07-25 01:46:57 -0700315 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700316 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700317 * read_s64() is a signed version of read_u64()
318 */
Paul Menagece16b492008-07-25 01:46:57 -0700319 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700320 /*
Paul Menage91796562008-04-29 01:00:01 -0700321 * read_map() is used for defining a map of key/value
322 * pairs. It should call cb->fill(cb, key, value) for each
323 * entry. The key/value pairs (and their ordering) should not
324 * change between reboots.
325 */
Paul Menagece16b492008-07-25 01:46:57 -0700326 int (*read_map)(struct cgroup *cont, struct cftype *cft,
327 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700328 /*
329 * read_seq_string() is used for outputting a simple sequence
330 * using seqfile.
331 */
Paul Menagece16b492008-07-25 01:46:57 -0700332 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
333 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700334
Paul Menagece16b492008-07-25 01:46:57 -0700335 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
336 struct file *file,
337 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700338
339 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700340 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700341 * a single integer (as parsed by simple_strtoull) from
342 * userspace. Use in place of write(); return 0 or error.
343 */
Paul Menagece16b492008-07-25 01:46:57 -0700344 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700345 /*
346 * write_s64() is a signed version of write_u64()
347 */
Paul Menagece16b492008-07-25 01:46:57 -0700348 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700349
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700350 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700351 * write_string() is passed a nul-terminated kernelspace
352 * buffer of maximum length determined by max_write_len.
353 * Returns 0 or -ve error code.
354 */
355 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
356 const char *buffer);
357 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700358 * trigger() callback can be used to get some kick from the
359 * userspace, when the actual string written is not important
360 * at all. The private field can be used to determine the
361 * kick type for multiplexing.
362 */
363 int (*trigger)(struct cgroup *cgrp, unsigned int event);
364
Paul Menagece16b492008-07-25 01:46:57 -0700365 int (*release)(struct inode *inode, struct file *file);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800366
367 /*
368 * register_event() callback will be used to add new userspace
369 * waiter for changes related to the cftype. Implement it if
370 * you want to provide this functionality. Use eventfd_signal()
371 * on eventfd to send notification to userspace.
372 */
373 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
374 struct eventfd_ctx *eventfd, const char *args);
375 /*
376 * unregister_event() callback will be called when userspace
377 * closes the eventfd or on cgroup removing.
378 * This callback must be implemented, if you want provide
379 * notification functionality.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800380 */
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700381 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800382 struct eventfd_ctx *eventfd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700383};
384
Tejun Heo8e3f6542012-04-01 12:09:55 -0700385/*
386 * cftype_sets describe cftypes belonging to a subsystem and are chained at
387 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
388 * terminated by zero length name.
389 */
390struct cftype_set {
391 struct list_head node; /* chained at subsys->cftsets */
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400392 struct cftype *cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700393};
394
Cliff Wickman31a7df02008-02-07 00:14:42 -0800395struct cgroup_scanner {
396 struct cgroup *cg;
397 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
398 void (*process_task)(struct task_struct *p,
399 struct cgroup_scanner *scan);
400 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700401 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800402};
403
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400404int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
405int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700406
Li Zefanffd2d882008-02-23 15:24:09 -0800407int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700408
Li Zefanffd2d882008-02-23 15:24:09 -0800409int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700410
Li Zefanffd2d882008-02-23 15:24:09 -0800411int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700412
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700413/* Return true if cgrp is a descendant of the task's cgroup */
414int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700415
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100416/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800417 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
418 * methods.
419 */
420struct cgroup_taskset;
421struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
422struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
423struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
424int cgroup_taskset_size(struct cgroup_taskset *tset);
425
426/**
427 * cgroup_taskset_for_each - iterate cgroup_taskset
428 * @task: the loop cursor
429 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
430 * @tset: taskset to iterate
431 */
432#define cgroup_taskset_for_each(task, skip_cgrp, tset) \
433 for ((task) = cgroup_taskset_first((tset)); (task); \
434 (task) = cgroup_taskset_next((tset))) \
435 if (!(skip_cgrp) || \
436 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
437
438/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100439 * Control Group subsystem type.
440 * See Documentation/cgroups/cgroups.txt for details
441 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700442
443struct cgroup_subsys {
Tejun Heo92fb9742012-11-19 08:13:38 -0800444 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
445 int (*css_online)(struct cgroup *cgrp);
446 void (*css_offline)(struct cgroup *cgrp);
447 void (*css_free)(struct cgroup *cgrp);
448
Li Zefan761b3ef2012-01-31 13:47:36 +0800449 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
450 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
451 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
452 void (*fork)(struct task_struct *task);
453 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
454 struct task_struct *task);
Li Zefan761b3ef2012-01-31 13:47:36 +0800455 void (*post_clone)(struct cgroup *cgrp);
456 void (*bind)(struct cgroup *root);
Hugh Dickinse5991372009-01-06 14:39:22 -0800457
Paul Menageddbcc7e2007-10-18 23:39:30 -0700458 int subsys_id;
459 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700460 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700461 int early_init;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700462 /*
463 * True if this subsys uses ID. ID is not available before cgroup_init()
464 * (not available in early_init time.)
465 */
466 bool use_id;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700467
468 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700469 * If %false, this subsystem is properly hierarchical -
470 * configuration, resource accounting and restriction on a parent
471 * cgroup cover those of its children. If %true, hierarchy support
472 * is broken in some ways - some subsystems ignore hierarchy
473 * completely while others are only implemented half-way.
474 *
475 * It's now disallowed to create nested cgroups if the subsystem is
476 * broken and cgroup core will emit a warning message on such
477 * cases. Eventually, all subsystems will be made properly
478 * hierarchical and this will go away.
479 */
480 bool broken_hierarchy;
481 bool warned_broken_hierarchy;
482
Paul Menageddbcc7e2007-10-18 23:39:30 -0700483#define MAX_CGROUP_TYPE_NAMELEN 32
484 const char *name;
485
Paul Menage999cd8a2009-01-07 18:08:36 -0800486 /*
Paul Menage999cd8a2009-01-07 18:08:36 -0800487 * Link to parent, and list entry in parent's children.
Li Zefan6be96a52012-06-06 19:12:30 -0700488 * Protected by cgroup_lock()
Paul Menage999cd8a2009-01-07 18:08:36 -0800489 */
490 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700491 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700492 /* used when use_id == true */
493 struct idr idr;
Hugh Dickins42aee6c2012-03-21 16:34:21 -0700494 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800495
Tejun Heo8e3f6542012-04-01 12:09:55 -0700496 /* list of cftype_sets */
497 struct list_head cftsets;
498
499 /* base cftypes, automatically [de]registered with subsys itself */
500 struct cftype *base_cftypes;
501 struct cftype_set base_cftset;
502
Ben Blume6a11052010-03-10 15:22:09 -0800503 /* should be defined only by modular subsystems */
504 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700505};
506
507#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200508#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700509#include <linux/cgroup_subsys.h>
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200510#undef IS_SUBSYS_ENABLED
Paul Menageddbcc7e2007-10-18 23:39:30 -0700511#undef SUBSYS
512
513static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800514 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700515{
Li Zefanffd2d882008-02-23 15:24:09 -0800516 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700517}
518
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200519/*
520 * function to get the cgroup_subsys_state which allows for extra
521 * rcu_dereference_check() conditions, such as locks used during the
522 * cgroup_subsys::attach() methods.
523 */
524#define task_subsys_state_check(task, subsys_id, __c) \
525 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200526 lockdep_is_held(&task->alloc_lock) || \
527 cgroup_lock_is_held() || (__c))
528
529static inline struct cgroup_subsys_state *
530task_subsys_state(struct task_struct *task, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700531{
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200532 return task_subsys_state_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700533}
534
535static inline struct cgroup* task_cgroup(struct task_struct *task,
536 int subsys_id)
537{
538 return task_subsys_state(task, subsys_id)->cgroup;
539}
540
Tejun Heo574bd9f2012-11-09 09:12:29 -0800541/**
542 * cgroup_for_each_child - iterate through children of a cgroup
543 * @pos: the cgroup * to use as the loop cursor
544 * @cgroup: cgroup whose children to walk
545 *
546 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
Tejun Heo92fb9742012-11-19 08:13:38 -0800547 * cgroup which hasn't finished ->css_online() or already has finished
548 * ->css_offline() may show up during traversal and it's each subsystem's
Tejun Heo574bd9f2012-11-09 09:12:29 -0800549 * responsibility to verify that each @pos is alive.
550 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800551 * If a subsystem synchronizes against the parent in its ->css_online() and
552 * before starting iterating, a cgroup which finished ->css_online() is
553 * guaranteed to be visible in the future iterations.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800554 */
555#define cgroup_for_each_child(pos, cgroup) \
556 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
557
558struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
559 struct cgroup *cgroup);
560
561/**
562 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
563 * @pos: the cgroup * to use as the loop cursor
564 * @cgroup: cgroup whose descendants to walk
565 *
566 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
Tejun Heo92fb9742012-11-19 08:13:38 -0800567 * descendant cgroup which hasn't finished ->css_online() or already has
568 * finished ->css_offline() may show up during traversal and it's each
Tejun Heo574bd9f2012-11-09 09:12:29 -0800569 * subsystem's responsibility to verify that each @pos is alive.
570 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800571 * If a subsystem synchronizes against the parent in its ->css_online() and
572 * before starting iterating, and synchronizes against @pos on each
573 * iteration, any descendant cgroup which finished ->css_offline() is
Tejun Heo574bd9f2012-11-09 09:12:29 -0800574 * guaranteed to be visible in the future iterations.
575 *
576 * In other words, the following guarantees that a descendant can't escape
577 * state updates of its ancestors.
578 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800579 * my_online(@cgrp)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800580 * {
581 * Lock @cgrp->parent and @cgrp;
582 * Inherit state from @cgrp->parent;
583 * Unlock both.
584 * }
585 *
586 * my_update_state(@cgrp)
587 * {
588 * Lock @cgrp;
589 * Update @cgrp's state;
590 * Unlock @cgrp;
591 *
592 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
593 * Lock @pos;
594 * Verify @pos is alive and inherit state from @pos->parent;
595 * Unlock @pos;
596 * }
597 * }
598 *
599 * As long as the inheriting step, including checking the parent state, is
600 * enclosed inside @pos locking, double-locking the parent isn't necessary
601 * while inheriting. The state update to the parent is guaranteed to be
602 * visible by walking order and, as long as inheriting operations to the
603 * same @pos are atomic to each other, multiple updates racing each other
604 * still result in the correct state. It's guaranateed that at least one
605 * inheritance happens for any cgroup after the latest update to its
606 * parent.
607 *
608 * If checking parent's state requires locking the parent, each inheriting
609 * iteration should lock and unlock both @pos->parent and @pos.
610 *
611 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800612 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800613 * operations.
614 */
615#define cgroup_for_each_descendant_pre(pos, cgroup) \
616 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
617 pos = cgroup_next_descendant_pre((pos), (cgroup)))
618
619struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
620 struct cgroup *cgroup);
621
622/**
623 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
624 * @pos: the cgroup * to use as the loop cursor
625 * @cgroup: cgroup whose descendants to walk
626 *
627 * Similar to cgroup_for_each_descendant_pre() but performs post-order
628 * traversal instead. Note that the walk visibility guarantee described in
629 * pre-order walk doesn't apply the same to post-order walks.
630 */
631#define cgroup_for_each_descendant_post(pos, cgroup) \
632 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
633 pos = cgroup_next_descendant_post((pos), (cgroup)))
634
Paul Menage817929e2007-10-18 23:39:36 -0700635/* A cgroup_iter should be treated as an opaque object */
636struct cgroup_iter {
637 struct list_head *cg_link;
638 struct list_head *task;
639};
640
Paul Menaged20a3902009-04-02 16:57:22 -0700641/*
642 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700643 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -0400644 * 1) call cgroup_iter_start to initialize an iterator
Paul Menage817929e2007-10-18 23:39:36 -0700645 *
646 * 2) call cgroup_iter_next() to retrieve member tasks until it
647 * returns NULL or until you want to end the iteration
648 *
649 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800650 *
Paul Menaged20a3902009-04-02 16:57:22 -0700651 * Or, call cgroup_scan_tasks() to iterate through every task in a
652 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
653 * the test_task() callback, but not while calling the process_task()
654 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700655 */
Li Zefanffd2d882008-02-23 15:24:09 -0800656void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
657struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700658 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800659void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800660int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800661int cgroup_attach_task(struct cgroup *, struct task_struct *);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700662int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
663
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700664/*
665 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
666 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
667 * CSS ID is assigned at cgroup allocation (create) automatically
668 * and removed when subsys calls free_css_id() function. This is because
669 * the lifetime of cgroup_subsys_state is subsys's matter.
670 *
671 * Looking up and scanning function should be called under rcu_read_lock().
Li Zefan6be96a52012-06-06 19:12:30 -0700672 * Taking cgroup_mutex is not necessary for following calls.
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700673 * But the css returned by this routine can be "not populated yet" or "being
674 * destroyed". The caller should check css and cgroup's status.
675 */
676
677/*
678 * Typically Called at ->destroy(), or somewhere the subsys frees
679 * cgroup_subsys_state.
680 */
681void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
682
683/* Find a cgroup_subsys_state which has given ID */
684
685struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
686
687/*
688 * Get a cgroup whose id is greater than or equal to id under tree of root.
689 * Returning a cgroup_subsys_state or NULL.
690 */
691struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
692 struct cgroup_subsys_state *root, int *foundid);
693
694/* Returns true if root is ancestor of cg */
695bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700696 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700697
698/* Get id and depth of css */
699unsigned short css_id(struct cgroup_subsys_state *css);
700unsigned short css_depth(struct cgroup_subsys_state *css);
Stephane Eraniane5d13672011-02-14 11:20:01 +0200701struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700702
Paul Menageddbcc7e2007-10-18 23:39:30 -0700703#else /* !CONFIG_CGROUPS */
704
705static inline int cgroup_init_early(void) { return 0; }
706static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700707static inline void cgroup_fork(struct task_struct *p) {}
708static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700709static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700710static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700711
712static inline void cgroup_lock(void) {}
713static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700714static inline int cgroupstats_build(struct cgroupstats *stats,
715 struct dentry *dentry)
716{
717 return -EINVAL;
718}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700719
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200720/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700721static inline int cgroup_attach_task_all(struct task_struct *from,
722 struct task_struct *t)
723{
724 return 0;
725}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200726
Paul Menageddbcc7e2007-10-18 23:39:30 -0700727#endif /* !CONFIG_CGROUPS */
728
Paul Menageddbcc7e2007-10-18 23:39:30 -0700729#endif /* _LINUX_CGROUP_H */