blob: 5f76829dd75e19f49b0c716d353dd8f3f8df34a7 [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
Li Zefan7d8e0bf2013-03-05 10:57:03 +080047/*
48 * Define the enumeration of all cgroup subsystems.
49 *
50 * We define ids for builtin subsystems and then modular ones.
51 */
Paul Menage817929e2007-10-18 23:39:36 -070052#define SUBSYS(_x) _x ## _subsys_id,
53enum cgroup_subsys_id {
Li Zefan7d8e0bf2013-03-05 10:57:03 +080054#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
Paul Menage817929e2007-10-18 23:39:36 -070055#include <linux/cgroup_subsys.h>
Li Zefan7d8e0bf2013-03-05 10:57:03 +080056#undef IS_SUBSYS_ENABLED
57 CGROUP_BUILTIN_SUBSYS_COUNT,
58
59 __CGROUP_SUBSYS_TEMP_PLACEHOLDER = CGROUP_BUILTIN_SUBSYS_COUNT - 1,
60
61#define IS_SUBSYS_ENABLED(option) IS_MODULE(option)
62#include <linux/cgroup_subsys.h>
63#undef IS_SUBSYS_ENABLED
Daniel Wagnera6f00292012-09-12 16:12:08 +020064 CGROUP_SUBSYS_COUNT,
Paul Menage817929e2007-10-18 23:39:36 -070065};
66#undef SUBSYS
67
Paul Menageddbcc7e2007-10-18 23:39:30 -070068/* Per-subsystem/per-cgroup state maintained by the system. */
69struct cgroup_subsys_state {
Paul Menaged20a3902009-04-02 16:57:22 -070070 /*
71 * The cgroup that this subsystem is attached to. Useful
Paul Menageddbcc7e2007-10-18 23:39:30 -070072 * for subsystems that want to know about the cgroup
Paul Menaged20a3902009-04-02 16:57:22 -070073 * hierarchy structure
74 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070075 struct cgroup *cgroup;
76
Paul Menaged20a3902009-04-02 16:57:22 -070077 /*
78 * State maintained by the cgroup system to allow subsystems
Paul Menagee7c5ec92009-01-07 18:08:38 -080079 * to be "busy". Should be accessed via css_get(),
Tao Mad0b2fdd2012-11-20 22:06:18 +080080 * css_tryget() and css_put().
Paul Menaged20a3902009-04-02 16:57:22 -070081 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070082
83 atomic_t refcnt;
84
85 unsigned long flags;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070086 /* ID for this css, if possible */
Arnd Bergmann2c392b82010-02-24 19:41:39 +010087 struct css_id __rcu *id;
Tejun Heo48ddbe12012-04-01 12:09:56 -070088
89 /* Used to put @cgroup->dentry on the last css_put() */
90 struct work_struct dput_work;
Paul Menageddbcc7e2007-10-18 23:39:30 -070091};
92
93/* bits in struct cgroup_subsys_state flags field */
94enum {
Tejun Heo38b53ab2012-11-19 08:13:36 -080095 CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */
Tejun Heo92fb9742012-11-19 08:13:38 -080096 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
Paul Menageddbcc7e2007-10-18 23:39:30 -070097};
98
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -080099/* Caller must verify that the css is not for root cgroup */
100static inline void __css_get(struct cgroup_subsys_state *css, int count)
101{
102 atomic_add(count, &css->refcnt);
103}
104
Paul Menageddbcc7e2007-10-18 23:39:30 -0700105/*
Paul Menagee7c5ec92009-01-07 18:08:38 -0800106 * Call css_get() to hold a reference on the css; it can be used
107 * for a reference obtained via:
108 * - an existing ref-counted reference to the css
109 * - task->cgroups for a locked task
Paul Menageddbcc7e2007-10-18 23:39:30 -0700110 */
111
112static inline void css_get(struct cgroup_subsys_state *css)
113{
114 /* We don't need to reference count the root state */
Tejun Heo38b53ab2012-11-19 08:13:36 -0800115 if (!(css->flags & CSS_ROOT))
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -0800116 __css_get(css, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700117}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800118
Paul Menagee7c5ec92009-01-07 18:08:38 -0800119/*
120 * Call css_tryget() to take a reference on a css if your existing
121 * (known-valid) reference isn't already ref-counted. Returns false if
122 * the css has been destroyed.
123 */
124
Tejun Heo28b4c272012-04-01 12:09:56 -0700125extern bool __css_tryget(struct cgroup_subsys_state *css);
Paul Menagee7c5ec92009-01-07 18:08:38 -0800126static inline bool css_tryget(struct cgroup_subsys_state *css)
127{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800128 if (css->flags & CSS_ROOT)
Paul Menagee7c5ec92009-01-07 18:08:38 -0800129 return true;
Tejun Heo28b4c272012-04-01 12:09:56 -0700130 return __css_tryget(css);
Paul Menagee7c5ec92009-01-07 18:08:38 -0800131}
132
Paul Menageddbcc7e2007-10-18 23:39:30 -0700133/*
134 * css_put() should be called to release a reference taken by
Paul Menagee7c5ec92009-01-07 18:08:38 -0800135 * css_get() or css_tryget()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700136 */
137
Tejun Heo28b4c272012-04-01 12:09:56 -0700138extern void __css_put(struct cgroup_subsys_state *css);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700139static inline void css_put(struct cgroup_subsys_state *css)
140{
Tejun Heo38b53ab2012-11-19 08:13:36 -0800141 if (!(css->flags & CSS_ROOT))
Tejun Heo28b4c272012-04-01 12:09:56 -0700142 __css_put(css);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700143}
144
Paul Menage3116f0e2008-04-29 01:00:04 -0700145/* bits in struct cgroup flags field */
146enum {
147 /* Control Group is dead */
148 CGRP_REMOVED,
Paul Menaged20a3902009-04-02 16:57:22 -0700149 /*
150 * Control Group has previously had a child cgroup or a task,
151 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
152 */
Paul Menage3116f0e2008-04-29 01:00:04 -0700153 CGRP_RELEASABLE,
154 /* Control Group requires release notifications to userspace */
155 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700156 /*
Tejun Heo2260e7f2012-11-19 08:13:38 -0800157 * Clone the parent's configuration when creating a new child
158 * cpuset cgroup. For historical reasons, this option can be
159 * specified at mount time and thus is implemented here.
Daniel Lezcano97978e62010-10-27 15:33:35 -0700160 */
Tejun Heo2260e7f2012-11-19 08:13:38 -0800161 CGRP_CPUSET_CLONE_CHILDREN,
Paul Menage3116f0e2008-04-29 01:00:04 -0700162};
163
Li Zefan65dff752013-03-01 15:01:56 +0800164struct cgroup_name {
165 struct rcu_head rcu_head;
166 char name[];
167};
168
Paul Menageddbcc7e2007-10-18 23:39:30 -0700169struct cgroup {
170 unsigned long flags; /* "unsigned long" so bitops work */
171
Paul Menaged20a3902009-04-02 16:57:22 -0700172 /*
173 * count users of this cgroup. >0 means busy, but doesn't
174 * necessarily indicate the number of tasks in the cgroup
175 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700176 atomic_t count;
177
Tejun Heo0a950f62012-11-19 09:02:12 -0800178 int id; /* ida allocated in-hierarchy ID */
179
Paul Menageddbcc7e2007-10-18 23:39:30 -0700180 /*
181 * We link our 'sibling' struct into our parent's 'children'.
182 * Our children link their 'sibling' into our 'children'.
183 */
184 struct list_head sibling; /* my parent's children */
185 struct list_head children; /* my children */
Tejun Heo05ef1d72012-04-01 12:09:56 -0700186 struct list_head files; /* my files */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700187
Paul Menaged20a3902009-04-02 16:57:22 -0700188 struct cgroup *parent; /* my parent */
Tejun Heofebfcef2012-11-19 08:13:36 -0800189 struct dentry *dentry; /* cgroup fs entry, RCU protected */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700190
Li Zefan65dff752013-03-01 15:01:56 +0800191 /*
192 * This is a copy of dentry->d_name, and it's needed because
193 * we can't use dentry->d_name in cgroup_path().
194 *
195 * You must acquire rcu_read_lock() to access cgrp->name, and
196 * the only place that can change it is rename(), which is
197 * protected by parent dir's i_mutex.
198 *
199 * Normally you should use cgroup_name() wrapper rather than
200 * access it directly.
201 */
202 struct cgroup_name __rcu *name;
203
Paul Menageddbcc7e2007-10-18 23:39:30 -0700204 /* Private pointers for each registered subsystem */
205 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
206
207 struct cgroupfs_root *root;
208 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700209
210 /*
211 * List of cg_cgroup_links pointing at css_sets with
212 * tasks in this cgroup. Protected by css_set_lock
213 */
214 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700215
Tejun Heob0ca5a82012-04-01 12:09:54 -0700216 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700217 struct list_head cft_q_node; /* used during cftype add/rm */
Tejun Heob0ca5a82012-04-01 12:09:54 -0700218
Paul Menage81a6a5c2007-10-18 23:39:38 -0700219 /*
220 * Linked list running through all cgroups that can
221 * potentially be reaped by the release agent. Protected by
222 * release_list_lock
223 */
224 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700225
Ben Blum72a8cb32009-09-23 15:56:27 -0700226 /*
227 * list of pidlists, up to two for each namespace (one for procs, one
228 * for tasks); created on demand.
229 */
230 struct list_head pidlists;
231 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800232
233 /* For RCU-protected deletion */
234 struct rcu_head rcu_head;
Li Zefanbe445622013-01-24 14:31:42 +0800235 struct work_struct free_work;
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800236
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300237 /* List of events which userspace want to receive */
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800238 struct list_head event_list;
239 spinlock_t event_list_lock;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400240
241 /* directory xattrs */
242 struct simple_xattrs xattrs;
Paul Menage817929e2007-10-18 23:39:36 -0700243};
244
Paul Menaged20a3902009-04-02 16:57:22 -0700245/*
246 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700247 * cgroup_subsys_state objects. This saves space in the task struct
248 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700249 * list_add()/del() can bump the reference count on the entire cgroup
250 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700251 */
252
253struct css_set {
254
255 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700256 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700257
258 /*
Li Zefan472b1052008-04-29 01:00:11 -0700259 * List running through all cgroup groups in the same hash
260 * slot. Protected by css_set_lock
261 */
262 struct hlist_node hlist;
263
264 /*
Paul Menage817929e2007-10-18 23:39:36 -0700265 * List running through all tasks using this cgroup
266 * group. Protected by css_set_lock
267 */
268 struct list_head tasks;
269
270 /*
271 * List of cg_cgroup_link objects on link chains from
272 * cgroups referenced from this css_set. Protected by
273 * css_set_lock
274 */
275 struct list_head cg_links;
276
277 /*
278 * Set of subsystem states, one for each subsystem. This array
279 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800280 * during subsystem registration (at boot time) and modular subsystem
281 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700282 */
283 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700284
285 /* For RCU-protected deletion */
286 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700287};
288
Paul Menage91796562008-04-29 01:00:01 -0700289/*
290 * cgroup_map_cb is an abstract callback API for reporting map-valued
291 * control files
292 */
293
294struct cgroup_map_cb {
295 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
296 void *state;
297};
298
Paul Menaged20a3902009-04-02 16:57:22 -0700299/*
300 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700301 *
302 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800303 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700304 * - the 'cftype' of the file is file->f_dentry->d_fsdata
305 */
306
Tejun Heo8e3f6542012-04-01 12:09:55 -0700307/* cftype->flags */
308#define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
Tao Mad0b2fdd2012-11-20 22:06:18 +0800309#define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700310
311#define MAX_CFTYPE_NAME 64
312
Paul Menageddbcc7e2007-10-18 23:39:30 -0700313struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700314 /*
315 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700316 * subsystem, followed by a period. Zero length string indicates
317 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700318 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700319 char name[MAX_CFTYPE_NAME];
320 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700321 /*
322 * If not 0, file mode is set to this value, otherwise it will
323 * be figured out automatically
324 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400325 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700326
327 /*
328 * If non-zero, defines the maximum length of string that can
329 * be passed to write_string; defaults to 64
330 */
331 size_t max_write_len;
332
Tejun Heo8e3f6542012-04-01 12:09:55 -0700333 /* CFTYPE_* flags */
334 unsigned int flags;
335
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400336 /* file xattrs */
337 struct simple_xattrs xattrs;
338
Paul Menagece16b492008-07-25 01:46:57 -0700339 int (*open)(struct inode *inode, struct file *file);
340 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
341 struct file *file,
342 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700343 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700344 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700345 * single integer. Use it in place of read()
346 */
Paul Menagece16b492008-07-25 01:46:57 -0700347 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700348 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700349 * read_s64() is a signed version of read_u64()
350 */
Paul Menagece16b492008-07-25 01:46:57 -0700351 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700352 /*
Paul Menage91796562008-04-29 01:00:01 -0700353 * read_map() is used for defining a map of key/value
354 * pairs. It should call cb->fill(cb, key, value) for each
355 * entry. The key/value pairs (and their ordering) should not
356 * change between reboots.
357 */
Paul Menagece16b492008-07-25 01:46:57 -0700358 int (*read_map)(struct cgroup *cont, struct cftype *cft,
359 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700360 /*
361 * read_seq_string() is used for outputting a simple sequence
362 * using seqfile.
363 */
Paul Menagece16b492008-07-25 01:46:57 -0700364 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
365 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700366
Paul Menagece16b492008-07-25 01:46:57 -0700367 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
368 struct file *file,
369 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700370
371 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700372 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700373 * a single integer (as parsed by simple_strtoull) from
374 * userspace. Use in place of write(); return 0 or error.
375 */
Paul Menagece16b492008-07-25 01:46:57 -0700376 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700377 /*
378 * write_s64() is a signed version of write_u64()
379 */
Paul Menagece16b492008-07-25 01:46:57 -0700380 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700381
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700382 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700383 * write_string() is passed a nul-terminated kernelspace
384 * buffer of maximum length determined by max_write_len.
385 * Returns 0 or -ve error code.
386 */
387 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
388 const char *buffer);
389 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700390 * trigger() callback can be used to get some kick from the
391 * userspace, when the actual string written is not important
392 * at all. The private field can be used to determine the
393 * kick type for multiplexing.
394 */
395 int (*trigger)(struct cgroup *cgrp, unsigned int event);
396
Paul Menagece16b492008-07-25 01:46:57 -0700397 int (*release)(struct inode *inode, struct file *file);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800398
399 /*
400 * register_event() callback will be used to add new userspace
401 * waiter for changes related to the cftype. Implement it if
402 * you want to provide this functionality. Use eventfd_signal()
403 * on eventfd to send notification to userspace.
404 */
405 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
406 struct eventfd_ctx *eventfd, const char *args);
407 /*
408 * unregister_event() callback will be called when userspace
409 * closes the eventfd or on cgroup removing.
410 * This callback must be implemented, if you want provide
411 * notification functionality.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800412 */
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700413 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800414 struct eventfd_ctx *eventfd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700415};
416
Tejun Heo8e3f6542012-04-01 12:09:55 -0700417/*
418 * cftype_sets describe cftypes belonging to a subsystem and are chained at
419 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
420 * terminated by zero length name.
421 */
422struct cftype_set {
423 struct list_head node; /* chained at subsys->cftsets */
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400424 struct cftype *cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700425};
426
Cliff Wickman31a7df02008-02-07 00:14:42 -0800427struct cgroup_scanner {
428 struct cgroup *cg;
429 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
430 void (*process_task)(struct task_struct *p,
431 struct cgroup_scanner *scan);
432 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700433 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800434};
435
Li Zefan65dff752013-03-01 15:01:56 +0800436/* Caller should hold rcu_read_lock() */
437static inline const char *cgroup_name(const struct cgroup *cgrp)
438{
439 return rcu_dereference(cgrp->name)->name;
440}
441
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400442int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
443int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700444
Li Zefanffd2d882008-02-23 15:24:09 -0800445int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700446
Li Zefanffd2d882008-02-23 15:24:09 -0800447int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700448
Li Zefanffd2d882008-02-23 15:24:09 -0800449int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700450
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700451/* Return true if cgrp is a descendant of the task's cgroup */
452int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700453
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100454/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800455 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
456 * methods.
457 */
458struct cgroup_taskset;
459struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
460struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
461struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
462int cgroup_taskset_size(struct cgroup_taskset *tset);
463
464/**
465 * cgroup_taskset_for_each - iterate cgroup_taskset
466 * @task: the loop cursor
467 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
468 * @tset: taskset to iterate
469 */
470#define cgroup_taskset_for_each(task, skip_cgrp, tset) \
471 for ((task) = cgroup_taskset_first((tset)); (task); \
472 (task) = cgroup_taskset_next((tset))) \
473 if (!(skip_cgrp) || \
474 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
475
476/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100477 * Control Group subsystem type.
478 * See Documentation/cgroups/cgroups.txt for details
479 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700480
481struct cgroup_subsys {
Tejun Heo92fb9742012-11-19 08:13:38 -0800482 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
483 int (*css_online)(struct cgroup *cgrp);
484 void (*css_offline)(struct cgroup *cgrp);
485 void (*css_free)(struct cgroup *cgrp);
486
Li Zefan761b3ef52012-01-31 13:47:36 +0800487 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
488 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
489 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
490 void (*fork)(struct task_struct *task);
491 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
492 struct task_struct *task);
Li Zefan761b3ef52012-01-31 13:47:36 +0800493 void (*bind)(struct cgroup *root);
Hugh Dickinse5991372009-01-06 14:39:22 -0800494
Paul Menageddbcc7e2007-10-18 23:39:30 -0700495 int subsys_id;
496 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700497 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700498 int early_init;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700499 /*
500 * True if this subsys uses ID. ID is not available before cgroup_init()
501 * (not available in early_init time.)
502 */
503 bool use_id;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700504
505 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700506 * If %false, this subsystem is properly hierarchical -
507 * configuration, resource accounting and restriction on a parent
508 * cgroup cover those of its children. If %true, hierarchy support
509 * is broken in some ways - some subsystems ignore hierarchy
510 * completely while others are only implemented half-way.
511 *
512 * It's now disallowed to create nested cgroups if the subsystem is
513 * broken and cgroup core will emit a warning message on such
514 * cases. Eventually, all subsystems will be made properly
515 * hierarchical and this will go away.
516 */
517 bool broken_hierarchy;
518 bool warned_broken_hierarchy;
519
Paul Menageddbcc7e2007-10-18 23:39:30 -0700520#define MAX_CGROUP_TYPE_NAMELEN 32
521 const char *name;
522
Paul Menage999cd8a2009-01-07 18:08:36 -0800523 /*
Paul Menage999cd8a2009-01-07 18:08:36 -0800524 * Link to parent, and list entry in parent's children.
Li Zefan6be96a52012-06-06 19:12:30 -0700525 * Protected by cgroup_lock()
Paul Menage999cd8a2009-01-07 18:08:36 -0800526 */
527 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700528 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700529 /* used when use_id == true */
530 struct idr idr;
Hugh Dickins42aee6c2012-03-21 16:34:21 -0700531 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800532
Tejun Heo8e3f6542012-04-01 12:09:55 -0700533 /* list of cftype_sets */
534 struct list_head cftsets;
535
536 /* base cftypes, automatically [de]registered with subsys itself */
537 struct cftype *base_cftypes;
538 struct cftype_set base_cftset;
539
Ben Blume6a11052010-03-10 15:22:09 -0800540 /* should be defined only by modular subsystems */
541 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700542};
543
544#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200545#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700546#include <linux/cgroup_subsys.h>
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200547#undef IS_SUBSYS_ENABLED
Paul Menageddbcc7e2007-10-18 23:39:30 -0700548#undef SUBSYS
549
550static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800551 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700552{
Li Zefanffd2d882008-02-23 15:24:09 -0800553 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700554}
555
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200556/*
557 * function to get the cgroup_subsys_state which allows for extra
558 * rcu_dereference_check() conditions, such as locks used during the
559 * cgroup_subsys::attach() methods.
560 */
561#define task_subsys_state_check(task, subsys_id, __c) \
562 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200563 lockdep_is_held(&task->alloc_lock) || \
564 cgroup_lock_is_held() || (__c))
565
566static inline struct cgroup_subsys_state *
567task_subsys_state(struct task_struct *task, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700568{
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200569 return task_subsys_state_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700570}
571
572static inline struct cgroup* task_cgroup(struct task_struct *task,
573 int subsys_id)
574{
575 return task_subsys_state(task, subsys_id)->cgroup;
576}
577
Tejun Heo574bd9f2012-11-09 09:12:29 -0800578/**
579 * cgroup_for_each_child - iterate through children of a cgroup
580 * @pos: the cgroup * to use as the loop cursor
581 * @cgroup: cgroup whose children to walk
582 *
583 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
Tejun Heo92fb9742012-11-19 08:13:38 -0800584 * cgroup which hasn't finished ->css_online() or already has finished
585 * ->css_offline() may show up during traversal and it's each subsystem's
Tejun Heo574bd9f2012-11-09 09:12:29 -0800586 * responsibility to verify that each @pos is alive.
587 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800588 * If a subsystem synchronizes against the parent in its ->css_online() and
589 * before starting iterating, a cgroup which finished ->css_online() is
590 * guaranteed to be visible in the future iterations.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800591 */
592#define cgroup_for_each_child(pos, cgroup) \
593 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
594
595struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
596 struct cgroup *cgroup);
Tejun Heo12a9d2f2013-01-07 08:49:33 -0800597struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
Tejun Heo574bd9f2012-11-09 09:12:29 -0800598
599/**
600 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
601 * @pos: the cgroup * to use as the loop cursor
602 * @cgroup: cgroup whose descendants to walk
603 *
604 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
Tejun Heo92fb9742012-11-19 08:13:38 -0800605 * descendant cgroup which hasn't finished ->css_online() or already has
606 * finished ->css_offline() may show up during traversal and it's each
Tejun Heo574bd9f2012-11-09 09:12:29 -0800607 * subsystem's responsibility to verify that each @pos is alive.
608 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800609 * If a subsystem synchronizes against the parent in its ->css_online() and
610 * before starting iterating, and synchronizes against @pos on each
611 * iteration, any descendant cgroup which finished ->css_offline() is
Tejun Heo574bd9f2012-11-09 09:12:29 -0800612 * guaranteed to be visible in the future iterations.
613 *
614 * In other words, the following guarantees that a descendant can't escape
615 * state updates of its ancestors.
616 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800617 * my_online(@cgrp)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800618 * {
619 * Lock @cgrp->parent and @cgrp;
620 * Inherit state from @cgrp->parent;
621 * Unlock both.
622 * }
623 *
624 * my_update_state(@cgrp)
625 * {
626 * Lock @cgrp;
627 * Update @cgrp's state;
628 * Unlock @cgrp;
629 *
630 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
631 * Lock @pos;
632 * Verify @pos is alive and inherit state from @pos->parent;
633 * Unlock @pos;
634 * }
635 * }
636 *
637 * As long as the inheriting step, including checking the parent state, is
638 * enclosed inside @pos locking, double-locking the parent isn't necessary
639 * while inheriting. The state update to the parent is guaranteed to be
640 * visible by walking order and, as long as inheriting operations to the
641 * same @pos are atomic to each other, multiple updates racing each other
642 * still result in the correct state. It's guaranateed that at least one
643 * inheritance happens for any cgroup after the latest update to its
644 * parent.
645 *
646 * If checking parent's state requires locking the parent, each inheriting
647 * iteration should lock and unlock both @pos->parent and @pos.
648 *
649 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800650 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800651 * operations.
652 */
653#define cgroup_for_each_descendant_pre(pos, cgroup) \
654 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
655 pos = cgroup_next_descendant_pre((pos), (cgroup)))
656
657struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
658 struct cgroup *cgroup);
659
660/**
661 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
662 * @pos: the cgroup * to use as the loop cursor
663 * @cgroup: cgroup whose descendants to walk
664 *
665 * Similar to cgroup_for_each_descendant_pre() but performs post-order
666 * traversal instead. Note that the walk visibility guarantee described in
667 * pre-order walk doesn't apply the same to post-order walks.
668 */
669#define cgroup_for_each_descendant_post(pos, cgroup) \
670 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
671 pos = cgroup_next_descendant_post((pos), (cgroup)))
672
Paul Menage817929e2007-10-18 23:39:36 -0700673/* A cgroup_iter should be treated as an opaque object */
674struct cgroup_iter {
675 struct list_head *cg_link;
676 struct list_head *task;
677};
678
Paul Menaged20a3902009-04-02 16:57:22 -0700679/*
680 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700681 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -0400682 * 1) call cgroup_iter_start to initialize an iterator
Paul Menage817929e2007-10-18 23:39:36 -0700683 *
684 * 2) call cgroup_iter_next() to retrieve member tasks until it
685 * returns NULL or until you want to end the iteration
686 *
687 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800688 *
Paul Menaged20a3902009-04-02 16:57:22 -0700689 * Or, call cgroup_scan_tasks() to iterate through every task in a
690 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
691 * the test_task() callback, but not while calling the process_task()
692 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700693 */
Li Zefanffd2d882008-02-23 15:24:09 -0800694void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
695struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700696 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800697void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800698int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800699int cgroup_attach_task(struct cgroup *, struct task_struct *);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700700int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
701
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700702/*
703 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
704 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
705 * CSS ID is assigned at cgroup allocation (create) automatically
706 * and removed when subsys calls free_css_id() function. This is because
707 * the lifetime of cgroup_subsys_state is subsys's matter.
708 *
709 * Looking up and scanning function should be called under rcu_read_lock().
Li Zefan6be96a52012-06-06 19:12:30 -0700710 * Taking cgroup_mutex is not necessary for following calls.
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700711 * But the css returned by this routine can be "not populated yet" or "being
712 * destroyed". The caller should check css and cgroup's status.
713 */
714
715/*
716 * Typically Called at ->destroy(), or somewhere the subsys frees
717 * cgroup_subsys_state.
718 */
719void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
720
721/* Find a cgroup_subsys_state which has given ID */
722
723struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
724
725/*
726 * Get a cgroup whose id is greater than or equal to id under tree of root.
727 * Returning a cgroup_subsys_state or NULL.
728 */
729struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
730 struct cgroup_subsys_state *root, int *foundid);
731
732/* Returns true if root is ancestor of cg */
733bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700734 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700735
736/* Get id and depth of css */
737unsigned short css_id(struct cgroup_subsys_state *css);
738unsigned short css_depth(struct cgroup_subsys_state *css);
Stephane Eraniane5d13672011-02-14 11:20:01 +0200739struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700740
Paul Menageddbcc7e2007-10-18 23:39:30 -0700741#else /* !CONFIG_CGROUPS */
742
743static inline int cgroup_init_early(void) { return 0; }
744static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700745static inline void cgroup_fork(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700746static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700747static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700748
749static inline void cgroup_lock(void) {}
750static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700751static inline int cgroupstats_build(struct cgroupstats *stats,
752 struct dentry *dentry)
753{
754 return -EINVAL;
755}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700756
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200757/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700758static inline int cgroup_attach_task_all(struct task_struct *from,
759 struct task_struct *t)
760{
761 return 0;
762}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200763
Paul Menageddbcc7e2007-10-18 23:39:30 -0700764#endif /* !CONFIG_CGROUPS */
765
Paul Menageddbcc7e2007-10-18 23:39:30 -0700766#endif /* _LINUX_CGROUP_H */