blob: 7d73905dcba2166f91fd029518389ab0bbe66d1f [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(),
Tao Mad0b2fdd2012-11-20 22:06:18 +080069 * css_tryget() and css_put().
Paul Menaged20a3902009-04-02 16:57:22 -070070 */
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
Tejun Heo0a950f62012-11-19 09:02:12 -0800162 int id; /* ida allocated in-hierarchy ID */
163
Paul Menageddbcc7e2007-10-18 23:39:30 -0700164 /*
165 * We link our 'sibling' struct into our parent's 'children'.
166 * Our children link their 'sibling' into our 'children'.
167 */
168 struct list_head sibling; /* my parent's children */
169 struct list_head children; /* my children */
Tejun Heo05ef1d72012-04-01 12:09:56 -0700170 struct list_head files; /* my files */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700171
Paul Menaged20a3902009-04-02 16:57:22 -0700172 struct cgroup *parent; /* my parent */
Tejun Heofebfcef2012-11-19 08:13:36 -0800173 struct dentry *dentry; /* cgroup fs entry, RCU protected */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700174
175 /* Private pointers for each registered subsystem */
176 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
177
178 struct cgroupfs_root *root;
179 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700180
181 /*
182 * List of cg_cgroup_links pointing at css_sets with
183 * tasks in this cgroup. Protected by css_set_lock
184 */
185 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700186
Tejun Heob0ca5a82012-04-01 12:09:54 -0700187 struct list_head allcg_node; /* cgroupfs_root->allcg_list */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700188 struct list_head cft_q_node; /* used during cftype add/rm */
Tejun Heob0ca5a82012-04-01 12:09:54 -0700189
Paul Menage81a6a5c2007-10-18 23:39:38 -0700190 /*
191 * Linked list running through all cgroups that can
192 * potentially be reaped by the release agent. Protected by
193 * release_list_lock
194 */
195 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700196
Ben Blum72a8cb32009-09-23 15:56:27 -0700197 /*
198 * list of pidlists, up to two for each namespace (one for procs, one
199 * for tasks); created on demand.
200 */
201 struct list_head pidlists;
202 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800203
204 /* For RCU-protected deletion */
205 struct rcu_head rcu_head;
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800206
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300207 /* List of events which userspace want to receive */
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800208 struct list_head event_list;
209 spinlock_t event_list_lock;
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400210
211 /* directory xattrs */
212 struct simple_xattrs xattrs;
Paul Menage817929e2007-10-18 23:39:36 -0700213};
214
Paul Menaged20a3902009-04-02 16:57:22 -0700215/*
216 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700217 * cgroup_subsys_state objects. This saves space in the task struct
218 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700219 * list_add()/del() can bump the reference count on the entire cgroup
220 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700221 */
222
223struct css_set {
224
225 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700226 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700227
228 /*
Li Zefan472b1052008-04-29 01:00:11 -0700229 * List running through all cgroup groups in the same hash
230 * slot. Protected by css_set_lock
231 */
232 struct hlist_node hlist;
233
234 /*
Paul Menage817929e2007-10-18 23:39:36 -0700235 * List running through all tasks using this cgroup
236 * group. Protected by css_set_lock
237 */
238 struct list_head tasks;
239
240 /*
241 * List of cg_cgroup_link objects on link chains from
242 * cgroups referenced from this css_set. Protected by
243 * css_set_lock
244 */
245 struct list_head cg_links;
246
247 /*
248 * Set of subsystem states, one for each subsystem. This array
249 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800250 * during subsystem registration (at boot time) and modular subsystem
251 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700252 */
253 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700254
255 /* For RCU-protected deletion */
256 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700257};
258
Paul Menage91796562008-04-29 01:00:01 -0700259/*
260 * cgroup_map_cb is an abstract callback API for reporting map-valued
261 * control files
262 */
263
264struct cgroup_map_cb {
265 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
266 void *state;
267};
268
Paul Menaged20a3902009-04-02 16:57:22 -0700269/*
270 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700271 *
272 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800273 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700274 * - the 'cftype' of the file is file->f_dentry->d_fsdata
275 */
276
Tejun Heo8e3f6542012-04-01 12:09:55 -0700277/* cftype->flags */
278#define CFTYPE_ONLY_ON_ROOT (1U << 0) /* only create on root cg */
Tao Mad0b2fdd2012-11-20 22:06:18 +0800279#define CFTYPE_NOT_ON_ROOT (1U << 1) /* don't create on root cg */
Tejun Heo8e3f6542012-04-01 12:09:55 -0700280
281#define MAX_CFTYPE_NAME 64
282
Paul Menageddbcc7e2007-10-18 23:39:30 -0700283struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700284 /*
285 * By convention, the name should begin with the name of the
Tejun Heo8e3f6542012-04-01 12:09:55 -0700286 * subsystem, followed by a period. Zero length string indicates
287 * end of cftype array.
Paul Menaged20a3902009-04-02 16:57:22 -0700288 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700289 char name[MAX_CFTYPE_NAME];
290 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700291 /*
292 * If not 0, file mode is set to this value, otherwise it will
293 * be figured out automatically
294 */
Al Viroa5e7ed32011-07-26 01:55:55 -0400295 umode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700296
297 /*
298 * If non-zero, defines the maximum length of string that can
299 * be passed to write_string; defaults to 64
300 */
301 size_t max_write_len;
302
Tejun Heo8e3f6542012-04-01 12:09:55 -0700303 /* CFTYPE_* flags */
304 unsigned int flags;
305
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400306 /* file xattrs */
307 struct simple_xattrs xattrs;
308
Paul Menagece16b492008-07-25 01:46:57 -0700309 int (*open)(struct inode *inode, struct file *file);
310 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
311 struct file *file,
312 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700313 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700314 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700315 * single integer. Use it in place of read()
316 */
Paul Menagece16b492008-07-25 01:46:57 -0700317 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700318 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700319 * read_s64() is a signed version of read_u64()
320 */
Paul Menagece16b492008-07-25 01:46:57 -0700321 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700322 /*
Paul Menage91796562008-04-29 01:00:01 -0700323 * read_map() is used for defining a map of key/value
324 * pairs. It should call cb->fill(cb, key, value) for each
325 * entry. The key/value pairs (and their ordering) should not
326 * change between reboots.
327 */
Paul Menagece16b492008-07-25 01:46:57 -0700328 int (*read_map)(struct cgroup *cont, struct cftype *cft,
329 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700330 /*
331 * read_seq_string() is used for outputting a simple sequence
332 * using seqfile.
333 */
Paul Menagece16b492008-07-25 01:46:57 -0700334 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
335 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700336
Paul Menagece16b492008-07-25 01:46:57 -0700337 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
338 struct file *file,
339 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700340
341 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700342 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700343 * a single integer (as parsed by simple_strtoull) from
344 * userspace. Use in place of write(); return 0 or error.
345 */
Paul Menagece16b492008-07-25 01:46:57 -0700346 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700347 /*
348 * write_s64() is a signed version of write_u64()
349 */
Paul Menagece16b492008-07-25 01:46:57 -0700350 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700351
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700352 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700353 * write_string() is passed a nul-terminated kernelspace
354 * buffer of maximum length determined by max_write_len.
355 * Returns 0 or -ve error code.
356 */
357 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
358 const char *buffer);
359 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700360 * trigger() callback can be used to get some kick from the
361 * userspace, when the actual string written is not important
362 * at all. The private field can be used to determine the
363 * kick type for multiplexing.
364 */
365 int (*trigger)(struct cgroup *cgrp, unsigned int event);
366
Paul Menagece16b492008-07-25 01:46:57 -0700367 int (*release)(struct inode *inode, struct file *file);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800368
369 /*
370 * register_event() callback will be used to add new userspace
371 * waiter for changes related to the cftype. Implement it if
372 * you want to provide this functionality. Use eventfd_signal()
373 * on eventfd to send notification to userspace.
374 */
375 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
376 struct eventfd_ctx *eventfd, const char *args);
377 /*
378 * unregister_event() callback will be called when userspace
379 * closes the eventfd or on cgroup removing.
380 * This callback must be implemented, if you want provide
381 * notification functionality.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800382 */
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700383 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800384 struct eventfd_ctx *eventfd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700385};
386
Tejun Heo8e3f6542012-04-01 12:09:55 -0700387/*
388 * cftype_sets describe cftypes belonging to a subsystem and are chained at
389 * cgroup_subsys->cftsets. Each cftset points to an array of cftypes
390 * terminated by zero length name.
391 */
392struct cftype_set {
393 struct list_head node; /* chained at subsys->cftsets */
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400394 struct cftype *cfts;
Tejun Heo8e3f6542012-04-01 12:09:55 -0700395};
396
Cliff Wickman31a7df02008-02-07 00:14:42 -0800397struct cgroup_scanner {
398 struct cgroup *cg;
399 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
400 void (*process_task)(struct task_struct *p,
401 struct cgroup_scanner *scan);
402 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700403 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800404};
405
Aristeu Rozanski03b1cde2012-08-23 16:53:30 -0400406int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
407int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
Tejun Heo8e3f6542012-04-01 12:09:55 -0700408
Li Zefanffd2d882008-02-23 15:24:09 -0800409int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700410
Li Zefanffd2d882008-02-23 15:24:09 -0800411int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700412
Li Zefanffd2d882008-02-23 15:24:09 -0800413int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700414
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700415/* Return true if cgrp is a descendant of the task's cgroup */
416int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700417
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100418/*
Tejun Heo2f7ee562011-12-12 18:12:21 -0800419 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
420 * methods.
421 */
422struct cgroup_taskset;
423struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
424struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
425struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset);
426int cgroup_taskset_size(struct cgroup_taskset *tset);
427
428/**
429 * cgroup_taskset_for_each - iterate cgroup_taskset
430 * @task: the loop cursor
431 * @skip_cgrp: skip if task's cgroup matches this, %NULL to iterate through all
432 * @tset: taskset to iterate
433 */
434#define cgroup_taskset_for_each(task, skip_cgrp, tset) \
435 for ((task) = cgroup_taskset_first((tset)); (task); \
436 (task) = cgroup_taskset_next((tset))) \
437 if (!(skip_cgrp) || \
438 cgroup_taskset_cur_cgroup((tset)) != (skip_cgrp))
439
440/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100441 * Control Group subsystem type.
442 * See Documentation/cgroups/cgroups.txt for details
443 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700444
445struct cgroup_subsys {
Tejun Heo92fb9742012-11-19 08:13:38 -0800446 struct cgroup_subsys_state *(*css_alloc)(struct cgroup *cgrp);
447 int (*css_online)(struct cgroup *cgrp);
448 void (*css_offline)(struct cgroup *cgrp);
449 void (*css_free)(struct cgroup *cgrp);
450
Li Zefan761b3ef2012-01-31 13:47:36 +0800451 int (*can_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
452 void (*cancel_attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
453 void (*attach)(struct cgroup *cgrp, struct cgroup_taskset *tset);
454 void (*fork)(struct task_struct *task);
455 void (*exit)(struct cgroup *cgrp, struct cgroup *old_cgrp,
456 struct task_struct *task);
Li Zefan761b3ef2012-01-31 13:47:36 +0800457 void (*bind)(struct cgroup *root);
Hugh Dickinse5991372009-01-06 14:39:22 -0800458
Paul Menageddbcc7e2007-10-18 23:39:30 -0700459 int subsys_id;
460 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700461 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700462 int early_init;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700463 /*
464 * True if this subsys uses ID. ID is not available before cgroup_init()
465 * (not available in early_init time.)
466 */
467 bool use_id;
Tejun Heo48ddbe12012-04-01 12:09:56 -0700468
469 /*
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700470 * If %false, this subsystem is properly hierarchical -
471 * configuration, resource accounting and restriction on a parent
472 * cgroup cover those of its children. If %true, hierarchy support
473 * is broken in some ways - some subsystems ignore hierarchy
474 * completely while others are only implemented half-way.
475 *
476 * It's now disallowed to create nested cgroups if the subsystem is
477 * broken and cgroup core will emit a warning message on such
478 * cases. Eventually, all subsystems will be made properly
479 * hierarchical and this will go away.
480 */
481 bool broken_hierarchy;
482 bool warned_broken_hierarchy;
483
Paul Menageddbcc7e2007-10-18 23:39:30 -0700484#define MAX_CGROUP_TYPE_NAMELEN 32
485 const char *name;
486
Paul Menage999cd8a2009-01-07 18:08:36 -0800487 /*
Paul Menage999cd8a2009-01-07 18:08:36 -0800488 * Link to parent, and list entry in parent's children.
Li Zefan6be96a52012-06-06 19:12:30 -0700489 * Protected by cgroup_lock()
Paul Menage999cd8a2009-01-07 18:08:36 -0800490 */
491 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700492 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700493 /* used when use_id == true */
494 struct idr idr;
Hugh Dickins42aee6c2012-03-21 16:34:21 -0700495 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800496
Tejun Heo8e3f6542012-04-01 12:09:55 -0700497 /* list of cftype_sets */
498 struct list_head cftsets;
499
500 /* base cftypes, automatically [de]registered with subsys itself */
501 struct cftype *base_cftypes;
502 struct cftype_set base_cftset;
503
Ben Blume6a11052010-03-10 15:22:09 -0800504 /* should be defined only by modular subsystems */
505 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700506};
507
508#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200509#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700510#include <linux/cgroup_subsys.h>
Daniel Wagner5fc0b022012-09-12 16:12:05 +0200511#undef IS_SUBSYS_ENABLED
Paul Menageddbcc7e2007-10-18 23:39:30 -0700512#undef SUBSYS
513
514static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800515 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700516{
Li Zefanffd2d882008-02-23 15:24:09 -0800517 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700518}
519
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200520/*
521 * function to get the cgroup_subsys_state which allows for extra
522 * rcu_dereference_check() conditions, such as locks used during the
523 * cgroup_subsys::attach() methods.
524 */
525#define task_subsys_state_check(task, subsys_id, __c) \
526 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200527 lockdep_is_held(&task->alloc_lock) || \
528 cgroup_lock_is_held() || (__c))
529
530static inline struct cgroup_subsys_state *
531task_subsys_state(struct task_struct *task, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700532{
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200533 return task_subsys_state_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700534}
535
536static inline struct cgroup* task_cgroup(struct task_struct *task,
537 int subsys_id)
538{
539 return task_subsys_state(task, subsys_id)->cgroup;
540}
541
Tejun Heo574bd9f2012-11-09 09:12:29 -0800542/**
543 * cgroup_for_each_child - iterate through children of a cgroup
544 * @pos: the cgroup * to use as the loop cursor
545 * @cgroup: cgroup whose children to walk
546 *
547 * Walk @cgroup's children. Must be called under rcu_read_lock(). A child
Tejun Heo92fb9742012-11-19 08:13:38 -0800548 * cgroup which hasn't finished ->css_online() or already has finished
549 * ->css_offline() may show up during traversal and it's each subsystem's
Tejun Heo574bd9f2012-11-09 09:12:29 -0800550 * responsibility to verify that each @pos is alive.
551 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800552 * If a subsystem synchronizes against the parent in its ->css_online() and
553 * before starting iterating, a cgroup which finished ->css_online() is
554 * guaranteed to be visible in the future iterations.
Tejun Heo574bd9f2012-11-09 09:12:29 -0800555 */
556#define cgroup_for_each_child(pos, cgroup) \
557 list_for_each_entry_rcu(pos, &(cgroup)->children, sibling)
558
559struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
560 struct cgroup *cgroup);
561
562/**
563 * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants
564 * @pos: the cgroup * to use as the loop cursor
565 * @cgroup: cgroup whose descendants to walk
566 *
567 * Walk @cgroup's descendants. Must be called under rcu_read_lock(). A
Tejun Heo92fb9742012-11-19 08:13:38 -0800568 * descendant cgroup which hasn't finished ->css_online() or already has
569 * finished ->css_offline() may show up during traversal and it's each
Tejun Heo574bd9f2012-11-09 09:12:29 -0800570 * subsystem's responsibility to verify that each @pos is alive.
571 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800572 * If a subsystem synchronizes against the parent in its ->css_online() and
573 * before starting iterating, and synchronizes against @pos on each
574 * iteration, any descendant cgroup which finished ->css_offline() is
Tejun Heo574bd9f2012-11-09 09:12:29 -0800575 * guaranteed to be visible in the future iterations.
576 *
577 * In other words, the following guarantees that a descendant can't escape
578 * state updates of its ancestors.
579 *
Tejun Heo92fb9742012-11-19 08:13:38 -0800580 * my_online(@cgrp)
Tejun Heo574bd9f2012-11-09 09:12:29 -0800581 * {
582 * Lock @cgrp->parent and @cgrp;
583 * Inherit state from @cgrp->parent;
584 * Unlock both.
585 * }
586 *
587 * my_update_state(@cgrp)
588 * {
589 * Lock @cgrp;
590 * Update @cgrp's state;
591 * Unlock @cgrp;
592 *
593 * cgroup_for_each_descendant_pre(@pos, @cgrp) {
594 * Lock @pos;
595 * Verify @pos is alive and inherit state from @pos->parent;
596 * Unlock @pos;
597 * }
598 * }
599 *
600 * As long as the inheriting step, including checking the parent state, is
601 * enclosed inside @pos locking, double-locking the parent isn't necessary
602 * while inheriting. The state update to the parent is guaranteed to be
603 * visible by walking order and, as long as inheriting operations to the
604 * same @pos are atomic to each other, multiple updates racing each other
605 * still result in the correct state. It's guaranateed that at least one
606 * inheritance happens for any cgroup after the latest update to its
607 * parent.
608 *
609 * If checking parent's state requires locking the parent, each inheriting
610 * iteration should lock and unlock both @pos->parent and @pos.
611 *
612 * Alternatively, a subsystem may choose to use a single global lock to
Tejun Heo92fb9742012-11-19 08:13:38 -0800613 * synchronize ->css_online() and ->css_offline() against tree-walking
Tejun Heo574bd9f2012-11-09 09:12:29 -0800614 * operations.
615 */
616#define cgroup_for_each_descendant_pre(pos, cgroup) \
617 for (pos = cgroup_next_descendant_pre(NULL, (cgroup)); (pos); \
618 pos = cgroup_next_descendant_pre((pos), (cgroup)))
619
620struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
621 struct cgroup *cgroup);
622
623/**
624 * cgroup_for_each_descendant_post - post-order walk of a cgroup's descendants
625 * @pos: the cgroup * to use as the loop cursor
626 * @cgroup: cgroup whose descendants to walk
627 *
628 * Similar to cgroup_for_each_descendant_pre() but performs post-order
629 * traversal instead. Note that the walk visibility guarantee described in
630 * pre-order walk doesn't apply the same to post-order walks.
631 */
632#define cgroup_for_each_descendant_post(pos, cgroup) \
633 for (pos = cgroup_next_descendant_post(NULL, (cgroup)); (pos); \
634 pos = cgroup_next_descendant_post((pos), (cgroup)))
635
Paul Menage817929e2007-10-18 23:39:36 -0700636/* A cgroup_iter should be treated as an opaque object */
637struct cgroup_iter {
638 struct list_head *cg_link;
639 struct list_head *task;
640};
641
Paul Menaged20a3902009-04-02 16:57:22 -0700642/*
643 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700644 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -0400645 * 1) call cgroup_iter_start to initialize an iterator
Paul Menage817929e2007-10-18 23:39:36 -0700646 *
647 * 2) call cgroup_iter_next() to retrieve member tasks until it
648 * returns NULL or until you want to end the iteration
649 *
650 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800651 *
Paul Menaged20a3902009-04-02 16:57:22 -0700652 * Or, call cgroup_scan_tasks() to iterate through every task in a
653 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
654 * the test_task() callback, but not while calling the process_task()
655 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700656 */
Li Zefanffd2d882008-02-23 15:24:09 -0800657void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
658struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700659 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800660void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800661int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800662int cgroup_attach_task(struct cgroup *, struct task_struct *);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700663int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
664
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700665/*
666 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
667 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
668 * CSS ID is assigned at cgroup allocation (create) automatically
669 * and removed when subsys calls free_css_id() function. This is because
670 * the lifetime of cgroup_subsys_state is subsys's matter.
671 *
672 * Looking up and scanning function should be called under rcu_read_lock().
Li Zefan6be96a52012-06-06 19:12:30 -0700673 * Taking cgroup_mutex is not necessary for following calls.
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700674 * But the css returned by this routine can be "not populated yet" or "being
675 * destroyed". The caller should check css and cgroup's status.
676 */
677
678/*
679 * Typically Called at ->destroy(), or somewhere the subsys frees
680 * cgroup_subsys_state.
681 */
682void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
683
684/* Find a cgroup_subsys_state which has given ID */
685
686struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
687
688/*
689 * Get a cgroup whose id is greater than or equal to id under tree of root.
690 * Returning a cgroup_subsys_state or NULL.
691 */
692struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
693 struct cgroup_subsys_state *root, int *foundid);
694
695/* Returns true if root is ancestor of cg */
696bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700697 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700698
699/* Get id and depth of css */
700unsigned short css_id(struct cgroup_subsys_state *css);
701unsigned short css_depth(struct cgroup_subsys_state *css);
Stephane Eraniane5d13672011-02-14 11:20:01 +0200702struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700703
Paul Menageddbcc7e2007-10-18 23:39:30 -0700704#else /* !CONFIG_CGROUPS */
705
706static inline int cgroup_init_early(void) { return 0; }
707static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700708static inline void cgroup_fork(struct task_struct *p) {}
709static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700710static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700711static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700712
713static inline void cgroup_lock(void) {}
714static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700715static inline int cgroupstats_build(struct cgroupstats *stats,
716 struct dentry *dentry)
717{
718 return -EINVAL;
719}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700720
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200721/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700722static inline int cgroup_attach_task_all(struct task_struct *from,
723 struct task_struct *t)
724{
725 return 0;
726}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200727
Paul Menageddbcc7e2007-10-18 23:39:30 -0700728#endif /* !CONFIG_CGROUPS */
729
Paul Menageddbcc7e2007-10-18 23:39:30 -0700730#endif /* _LINUX_CGROUP_H */