blob: 2a59d3101e5ddca3dc88f2b4fe472971a0ff40c9 [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>
Balbir Singh846c7bb2007-10-18 23:39:44 -070015#include <linux/cgroupstats.h>
Cliff Wickman31a7df02008-02-07 00:14:42 -080016#include <linux/prio_heap.h>
Paul Menagecc31edc2008-10-18 20:28:04 -070017#include <linux/rwsem.h>
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070018#include <linux/idr.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070019
20#ifdef CONFIG_CGROUPS
21
22struct cgroupfs_root;
23struct cgroup_subsys;
24struct inode;
Paul Menage84eea842008-07-25 01:47:00 -070025struct cgroup;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070026struct css_id;
Paul Menageddbcc7e2007-10-18 23:39:30 -070027
28extern int cgroup_init_early(void);
29extern int cgroup_init(void);
Paul Menageddbcc7e2007-10-18 23:39:30 -070030extern void cgroup_lock(void);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -080031extern int cgroup_lock_is_held(void);
Paul Menage84eea842008-07-25 01:47:00 -070032extern bool cgroup_lock_live_group(struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -070033extern void cgroup_unlock(void);
Paul Menageb4f48b62007-10-18 23:39:33 -070034extern void cgroup_fork(struct task_struct *p);
35extern void cgroup_fork_callbacks(struct task_struct *p);
Paul Menage817929e2007-10-18 23:39:36 -070036extern void cgroup_post_fork(struct task_struct *p);
Paul Menageb4f48b62007-10-18 23:39:33 -070037extern void cgroup_exit(struct task_struct *p, int run_callbacks);
Balbir Singh846c7bb2007-10-18 23:39:44 -070038extern int cgroupstats_build(struct cgroupstats *stats,
39 struct dentry *dentry);
Ben Blume6a11052010-03-10 15:22:09 -080040extern int cgroup_load_subsys(struct cgroup_subsys *ss);
Ben Blumcf5d5942010-03-10 15:22:09 -080041extern void cgroup_unload_subsys(struct cgroup_subsys *ss);
Paul Menageddbcc7e2007-10-18 23:39:30 -070042
Alexey Dobriyan828c0952009-10-01 15:43:56 -070043extern const struct file_operations proc_cgroup_operations;
Paul Menagea4243162007-10-18 23:39:35 -070044
Ben Blumaae8aab2010-03-10 15:22:07 -080045/* Define the enumeration of all builtin cgroup subsystems */
Paul Menage817929e2007-10-18 23:39:36 -070046#define SUBSYS(_x) _x ## _subsys_id,
47enum cgroup_subsys_id {
48#include <linux/cgroup_subsys.h>
Ben Blumaae8aab2010-03-10 15:22:07 -080049 CGROUP_BUILTIN_SUBSYS_COUNT
Paul Menage817929e2007-10-18 23:39:36 -070050};
51#undef SUBSYS
Ben Blumaae8aab2010-03-10 15:22:07 -080052/*
53 * This define indicates the maximum number of subsystems that can be loaded
54 * at once. We limit to this many since cgroupfs_root has subsys_bits to keep
55 * track of all of them.
56 */
57#define CGROUP_SUBSYS_COUNT (BITS_PER_BYTE*sizeof(unsigned long))
Paul Menage817929e2007-10-18 23:39:36 -070058
Paul Menageddbcc7e2007-10-18 23:39:30 -070059/* Per-subsystem/per-cgroup state maintained by the system. */
60struct cgroup_subsys_state {
Paul Menaged20a3902009-04-02 16:57:22 -070061 /*
62 * The cgroup that this subsystem is attached to. Useful
Paul Menageddbcc7e2007-10-18 23:39:30 -070063 * for subsystems that want to know about the cgroup
Paul Menaged20a3902009-04-02 16:57:22 -070064 * hierarchy structure
65 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070066 struct cgroup *cgroup;
67
Paul Menaged20a3902009-04-02 16:57:22 -070068 /*
69 * State maintained by the cgroup system to allow subsystems
Paul Menagee7c5ec92009-01-07 18:08:38 -080070 * to be "busy". Should be accessed via css_get(),
Paul Menaged20a3902009-04-02 16:57:22 -070071 * css_tryget() and and css_put().
72 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070073
74 atomic_t refcnt;
75
76 unsigned long flags;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -070077 /* ID for this css, if possible */
78 struct css_id *id;
Paul Menageddbcc7e2007-10-18 23:39:30 -070079};
80
81/* bits in struct cgroup_subsys_state flags field */
82enum {
83 CSS_ROOT, /* This CSS is the root of the subsystem */
Paul Menagee7c5ec92009-01-07 18:08:38 -080084 CSS_REMOVED, /* This CSS is dead */
Paul Menageddbcc7e2007-10-18 23:39:30 -070085};
86
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -080087/* Caller must verify that the css is not for root cgroup */
88static inline void __css_get(struct cgroup_subsys_state *css, int count)
89{
90 atomic_add(count, &css->refcnt);
91}
92
Paul Menageddbcc7e2007-10-18 23:39:30 -070093/*
Paul Menagee7c5ec92009-01-07 18:08:38 -080094 * Call css_get() to hold a reference on the css; it can be used
95 * for a reference obtained via:
96 * - an existing ref-counted reference to the css
97 * - task->cgroups for a locked task
Paul Menageddbcc7e2007-10-18 23:39:30 -070098 */
99
100static inline void css_get(struct cgroup_subsys_state *css)
101{
102 /* We don't need to reference count the root state */
103 if (!test_bit(CSS_ROOT, &css->flags))
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -0800104 __css_get(css, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700105}
Paul Menagee7c5ec92009-01-07 18:08:38 -0800106
107static inline bool css_is_removed(struct cgroup_subsys_state *css)
108{
109 return test_bit(CSS_REMOVED, &css->flags);
110}
111
112/*
113 * Call css_tryget() to take a reference on a css if your existing
114 * (known-valid) reference isn't already ref-counted. Returns false if
115 * the css has been destroyed.
116 */
117
118static inline bool css_tryget(struct cgroup_subsys_state *css)
119{
120 if (test_bit(CSS_ROOT, &css->flags))
121 return true;
122 while (!atomic_inc_not_zero(&css->refcnt)) {
123 if (test_bit(CSS_REMOVED, &css->flags))
124 return false;
Paul Menage804b3c22009-01-29 14:25:21 -0800125 cpu_relax();
Paul Menagee7c5ec92009-01-07 18:08:38 -0800126 }
127 return true;
128}
129
Paul Menageddbcc7e2007-10-18 23:39:30 -0700130/*
131 * css_put() should be called to release a reference taken by
Paul Menagee7c5ec92009-01-07 18:08:38 -0800132 * css_get() or css_tryget()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700133 */
134
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -0800135extern void __css_put(struct cgroup_subsys_state *css, int count);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700136static inline void css_put(struct cgroup_subsys_state *css)
137{
138 if (!test_bit(CSS_ROOT, &css->flags))
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -0800139 __css_put(css, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700140}
141
Paul Menage3116f0e2008-04-29 01:00:04 -0700142/* bits in struct cgroup flags field */
143enum {
144 /* Control Group is dead */
145 CGRP_REMOVED,
Paul Menaged20a3902009-04-02 16:57:22 -0700146 /*
147 * Control Group has previously had a child cgroup or a task,
148 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set)
149 */
Paul Menage3116f0e2008-04-29 01:00:04 -0700150 CGRP_RELEASABLE,
151 /* Control Group requires release notifications to userspace */
152 CGRP_NOTIFY_ON_RELEASE,
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700153 /*
154 * A thread in rmdir() is wating for this cgroup.
155 */
156 CGRP_WAIT_ON_RMDIR,
Paul Menage3116f0e2008-04-29 01:00:04 -0700157};
158
Ben Blum72a8cb32009-09-23 15:56:27 -0700159/* which pidlist file are we talking about? */
160enum cgroup_filetype {
161 CGROUP_FILE_PROCS,
162 CGROUP_FILE_TASKS,
163};
164
165/*
166 * A pidlist is a list of pids that virtually represents the contents of one
167 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
168 * a pair (one each for procs, tasks) for each pid namespace that's relevant
169 * to the cgroup.
170 */
Ben Blum102a7752009-09-23 15:56:26 -0700171struct cgroup_pidlist {
Ben Blum72a8cb32009-09-23 15:56:27 -0700172 /*
173 * used to find which pidlist is wanted. doesn't change as long as
174 * this particular list stays in the list.
175 */
176 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
Ben Blum102a7752009-09-23 15:56:26 -0700177 /* array of xids */
178 pid_t *list;
179 /* how many elements the above list has */
180 int length;
181 /* how many files are using the current array */
182 int use_count;
Ben Blum72a8cb32009-09-23 15:56:27 -0700183 /* each of these stored in a list by its cgroup */
184 struct list_head links;
185 /* pointer to the cgroup we belong to, for list removal purposes */
186 struct cgroup *owner;
187 /* protects the other fields */
188 struct rw_semaphore mutex;
Ben Blum102a7752009-09-23 15:56:26 -0700189};
190
Paul Menageddbcc7e2007-10-18 23:39:30 -0700191struct cgroup {
192 unsigned long flags; /* "unsigned long" so bitops work */
193
Paul Menaged20a3902009-04-02 16:57:22 -0700194 /*
195 * count users of this cgroup. >0 means busy, but doesn't
196 * necessarily indicate the number of tasks in the cgroup
197 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700198 atomic_t count;
199
200 /*
201 * We link our 'sibling' struct into our parent's 'children'.
202 * Our children link their 'sibling' into our 'children'.
203 */
204 struct list_head sibling; /* my parent's children */
205 struct list_head children; /* my children */
206
Paul Menaged20a3902009-04-02 16:57:22 -0700207 struct cgroup *parent; /* my parent */
Paul Menagea47295e2009-01-07 18:07:44 -0800208 struct dentry *dentry; /* cgroup fs entry, RCU protected */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700209
210 /* Private pointers for each registered subsystem */
211 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
212
213 struct cgroupfs_root *root;
214 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700215
216 /*
217 * List of cg_cgroup_links pointing at css_sets with
218 * tasks in this cgroup. Protected by css_set_lock
219 */
220 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700221
222 /*
223 * Linked list running through all cgroups that can
224 * potentially be reaped by the release agent. Protected by
225 * release_list_lock
226 */
227 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700228
Ben Blum72a8cb32009-09-23 15:56:27 -0700229 /*
230 * list of pidlists, up to two for each namespace (one for procs, one
231 * for tasks); created on demand.
232 */
233 struct list_head pidlists;
234 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800235
236 /* For RCU-protected deletion */
237 struct rcu_head rcu_head;
Paul Menage817929e2007-10-18 23:39:36 -0700238};
239
Paul Menaged20a3902009-04-02 16:57:22 -0700240/*
241 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700242 * cgroup_subsys_state objects. This saves space in the task struct
243 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700244 * list_add()/del() can bump the reference count on the entire cgroup
245 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700246 */
247
248struct css_set {
249
250 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700251 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700252
253 /*
Li Zefan472b1052008-04-29 01:00:11 -0700254 * List running through all cgroup groups in the same hash
255 * slot. Protected by css_set_lock
256 */
257 struct hlist_node hlist;
258
259 /*
Paul Menage817929e2007-10-18 23:39:36 -0700260 * List running through all tasks using this cgroup
261 * group. Protected by css_set_lock
262 */
263 struct list_head tasks;
264
265 /*
266 * List of cg_cgroup_link objects on link chains from
267 * cgroups referenced from this css_set. Protected by
268 * css_set_lock
269 */
270 struct list_head cg_links;
271
272 /*
273 * Set of subsystem states, one for each subsystem. This array
274 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800275 * during subsystem registration (at boot time) and modular subsystem
276 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700277 */
278 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700279
280 /* For RCU-protected deletion */
281 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700282};
283
Paul Menage91796562008-04-29 01:00:01 -0700284/*
285 * cgroup_map_cb is an abstract callback API for reporting map-valued
286 * control files
287 */
288
289struct cgroup_map_cb {
290 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
291 void *state;
292};
293
Paul Menaged20a3902009-04-02 16:57:22 -0700294/*
295 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700296 *
297 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800298 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700299 * - the 'cftype' of the file is file->f_dentry->d_fsdata
300 */
301
302#define MAX_CFTYPE_NAME 64
303struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700304 /*
305 * By convention, the name should begin with the name of the
306 * subsystem, followed by a period
307 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700308 char name[MAX_CFTYPE_NAME];
309 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700310 /*
311 * If not 0, file mode is set to this value, otherwise it will
312 * be figured out automatically
313 */
314 mode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700315
316 /*
317 * If non-zero, defines the maximum length of string that can
318 * be passed to write_string; defaults to 64
319 */
320 size_t max_write_len;
321
Paul Menagece16b492008-07-25 01:46:57 -0700322 int (*open)(struct inode *inode, struct file *file);
323 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
324 struct file *file,
325 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700326 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700327 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700328 * single integer. Use it in place of read()
329 */
Paul Menagece16b492008-07-25 01:46:57 -0700330 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700331 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700332 * read_s64() is a signed version of read_u64()
333 */
Paul Menagece16b492008-07-25 01:46:57 -0700334 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700335 /*
Paul Menage91796562008-04-29 01:00:01 -0700336 * read_map() is used for defining a map of key/value
337 * pairs. It should call cb->fill(cb, key, value) for each
338 * entry. The key/value pairs (and their ordering) should not
339 * change between reboots.
340 */
Paul Menagece16b492008-07-25 01:46:57 -0700341 int (*read_map)(struct cgroup *cont, struct cftype *cft,
342 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700343 /*
344 * read_seq_string() is used for outputting a simple sequence
345 * using seqfile.
346 */
Paul Menagece16b492008-07-25 01:46:57 -0700347 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
348 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700349
Paul Menagece16b492008-07-25 01:46:57 -0700350 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
351 struct file *file,
352 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700353
354 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700355 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700356 * a single integer (as parsed by simple_strtoull) from
357 * userspace. Use in place of write(); return 0 or error.
358 */
Paul Menagece16b492008-07-25 01:46:57 -0700359 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700360 /*
361 * write_s64() is a signed version of write_u64()
362 */
Paul Menagece16b492008-07-25 01:46:57 -0700363 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700364
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700365 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700366 * write_string() is passed a nul-terminated kernelspace
367 * buffer of maximum length determined by max_write_len.
368 * Returns 0 or -ve error code.
369 */
370 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
371 const char *buffer);
372 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700373 * trigger() callback can be used to get some kick from the
374 * userspace, when the actual string written is not important
375 * at all. The private field can be used to determine the
376 * kick type for multiplexing.
377 */
378 int (*trigger)(struct cgroup *cgrp, unsigned int event);
379
Paul Menagece16b492008-07-25 01:46:57 -0700380 int (*release)(struct inode *inode, struct file *file);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700381};
382
Cliff Wickman31a7df02008-02-07 00:14:42 -0800383struct cgroup_scanner {
384 struct cgroup *cg;
385 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
386 void (*process_task)(struct task_struct *p,
387 struct cgroup_scanner *scan);
388 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700389 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800390};
391
Paul Menaged20a3902009-04-02 16:57:22 -0700392/*
393 * Add a new file to the given cgroup directory. Should only be
394 * called by subsystems from within a populate() method
395 */
Li Zefanffd2d882008-02-23 15:24:09 -0800396int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700397 const struct cftype *cft);
398
Paul Menaged20a3902009-04-02 16:57:22 -0700399/*
400 * Add a set of new files to the given cgroup directory. Should
401 * only be called by subsystems from within a populate() method
402 */
Li Zefanffd2d882008-02-23 15:24:09 -0800403int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700404 struct cgroup_subsys *subsys,
405 const struct cftype cft[],
406 int count);
407
Li Zefanffd2d882008-02-23 15:24:09 -0800408int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700409
Li Zefanffd2d882008-02-23 15:24:09 -0800410int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700411
Li Zefanffd2d882008-02-23 15:24:09 -0800412int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700413
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700414/* Return true if cgrp is a descendant of the task's cgroup */
415int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700416
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100417/*
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700418 * When the subsys has to access css and may add permanent refcnt to css,
419 * it should take care of racy conditions with rmdir(). Following set of
420 * functions, is for stop/restart rmdir if necessary.
421 * Because these will call css_get/put, "css" should be alive css.
422 *
423 * cgroup_exclude_rmdir();
424 * ...do some jobs which may access arbitrary empty cgroup
425 * cgroup_release_and_wakeup_rmdir();
426 *
427 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
428 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
429 */
430
431void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
432void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
433
434/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100435 * Control Group subsystem type.
436 * See Documentation/cgroups/cgroups.txt for details
437 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700438
439struct cgroup_subsys {
440 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800441 struct cgroup *cgrp);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700442 int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Li Zefanffd2d882008-02-23 15:24:09 -0800443 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Ben Blumbe367d02009-09-23 15:56:31 -0700444 int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
445 struct task_struct *tsk, bool threadgroup);
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800446 void (*cancel_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
447 struct task_struct *tsk, bool threadgroup);
Li Zefanffd2d882008-02-23 15:24:09 -0800448 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -0700449 struct cgroup *old_cgrp, struct task_struct *tsk,
450 bool threadgroup);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700451 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
452 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
453 int (*populate)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800454 struct cgroup *cgrp);
455 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700456 void (*bind)(struct cgroup_subsys *ss, 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;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700467#define MAX_CGROUP_TYPE_NAMELEN 32
468 const char *name;
469
Paul Menage999cd8a2009-01-07 18:08:36 -0800470 /*
471 * Protects sibling/children links of cgroups in this
472 * hierarchy, plus protects which hierarchy (or none) the
473 * subsystem is a part of (i.e. root/sibling). To avoid
474 * potential deadlocks, the following operations should not be
475 * undertaken while holding any hierarchy_mutex:
476 *
477 * - allocating memory
478 * - initiating hotplug events
479 */
480 struct mutex hierarchy_mutex;
Li Zefancfebe562009-02-11 13:04:36 -0800481 struct lock_class_key subsys_key;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700482
Paul Menage999cd8a2009-01-07 18:08:36 -0800483 /*
484 * Link to parent, and list entry in parent's children.
485 * Protected by this->hierarchy_mutex and cgroup_lock()
486 */
487 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700488 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700489 /* used when use_id == true */
490 struct idr idr;
491 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800492
493 /* should be defined only by modular subsystems */
494 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700495};
496
497#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
498#include <linux/cgroup_subsys.h>
499#undef SUBSYS
500
501static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800502 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700503{
Li Zefanffd2d882008-02-23 15:24:09 -0800504 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700505}
506
507static inline struct cgroup_subsys_state *task_subsys_state(
508 struct task_struct *task, int subsys_id)
509{
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800510 return rcu_dereference_check(task->cgroups->subsys[subsys_id],
511 rcu_read_lock_held() ||
512 cgroup_lock_is_held());
Paul Menageddbcc7e2007-10-18 23:39:30 -0700513}
514
515static inline struct cgroup* task_cgroup(struct task_struct *task,
516 int subsys_id)
517{
518 return task_subsys_state(task, subsys_id)->cgroup;
519}
520
Serge E. Hallyne885dcd2008-07-25 01:47:06 -0700521int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
522 char *nodename);
Paul Menage697f4162007-10-18 23:39:34 -0700523
Paul Menage817929e2007-10-18 23:39:36 -0700524/* A cgroup_iter should be treated as an opaque object */
525struct cgroup_iter {
526 struct list_head *cg_link;
527 struct list_head *task;
528};
529
Paul Menaged20a3902009-04-02 16:57:22 -0700530/*
531 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700532 *
533 * 1) call cgroup_iter_start to intialize an iterator
534 *
535 * 2) call cgroup_iter_next() to retrieve member tasks until it
536 * returns NULL or until you want to end the iteration
537 *
538 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800539 *
Paul Menaged20a3902009-04-02 16:57:22 -0700540 * Or, call cgroup_scan_tasks() to iterate through every task in a
541 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
542 * the test_task() callback, but not while calling the process_task()
543 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700544 */
Li Zefanffd2d882008-02-23 15:24:09 -0800545void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
546struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700547 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800548void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800549int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800550int cgroup_attach_task(struct cgroup *, struct task_struct *);
Paul Menage817929e2007-10-18 23:39:36 -0700551
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700552/*
553 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
554 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
555 * CSS ID is assigned at cgroup allocation (create) automatically
556 * and removed when subsys calls free_css_id() function. This is because
557 * the lifetime of cgroup_subsys_state is subsys's matter.
558 *
559 * Looking up and scanning function should be called under rcu_read_lock().
560 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
561 * But the css returned by this routine can be "not populated yet" or "being
562 * destroyed". The caller should check css and cgroup's status.
563 */
564
565/*
566 * Typically Called at ->destroy(), or somewhere the subsys frees
567 * cgroup_subsys_state.
568 */
569void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
570
571/* Find a cgroup_subsys_state which has given ID */
572
573struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
574
575/*
576 * Get a cgroup whose id is greater than or equal to id under tree of root.
577 * Returning a cgroup_subsys_state or NULL.
578 */
579struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
580 struct cgroup_subsys_state *root, int *foundid);
581
582/* Returns true if root is ancestor of cg */
583bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700584 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700585
586/* Get id and depth of css */
587unsigned short css_id(struct cgroup_subsys_state *css);
588unsigned short css_depth(struct cgroup_subsys_state *css);
589
Paul Menageddbcc7e2007-10-18 23:39:30 -0700590#else /* !CONFIG_CGROUPS */
591
592static inline int cgroup_init_early(void) { return 0; }
593static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700594static inline void cgroup_fork(struct task_struct *p) {}
595static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700596static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700597static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700598
599static inline void cgroup_lock(void) {}
600static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700601static inline int cgroupstats_build(struct cgroupstats *stats,
602 struct dentry *dentry)
603{
604 return -EINVAL;
605}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700606
607#endif /* !CONFIG_CGROUPS */
608
Paul Menageddbcc7e2007-10-18 23:39:30 -0700609#endif /* _LINUX_CGROUP_H */