blob: ed4ba111bc8d32a34ff7c6558589f1d3dc7d073f [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 */
Arnd Bergmann2c392b82010-02-24 19:41:39 +010078 struct css_id __rcu *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,
Daniel Lezcano97978e62010-10-27 15:33:35 -0700157 /*
158 * Clone cgroup values when creating a new child cgroup
159 */
160 CGRP_CLONE_CHILDREN,
Paul Menage3116f0e2008-04-29 01:00:04 -0700161};
162
Ben Blum72a8cb32009-09-23 15:56:27 -0700163/* which pidlist file are we talking about? */
164enum cgroup_filetype {
165 CGROUP_FILE_PROCS,
166 CGROUP_FILE_TASKS,
167};
168
169/*
170 * A pidlist is a list of pids that virtually represents the contents of one
171 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
172 * a pair (one each for procs, tasks) for each pid namespace that's relevant
173 * to the cgroup.
174 */
Ben Blum102a7752009-09-23 15:56:26 -0700175struct cgroup_pidlist {
Ben Blum72a8cb32009-09-23 15:56:27 -0700176 /*
177 * used to find which pidlist is wanted. doesn't change as long as
178 * this particular list stays in the list.
179 */
180 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
Ben Blum102a7752009-09-23 15:56:26 -0700181 /* array of xids */
182 pid_t *list;
183 /* how many elements the above list has */
184 int length;
185 /* how many files are using the current array */
186 int use_count;
Ben Blum72a8cb32009-09-23 15:56:27 -0700187 /* each of these stored in a list by its cgroup */
188 struct list_head links;
189 /* pointer to the cgroup we belong to, for list removal purposes */
190 struct cgroup *owner;
191 /* protects the other fields */
192 struct rw_semaphore mutex;
Ben Blum102a7752009-09-23 15:56:26 -0700193};
194
Paul Menageddbcc7e2007-10-18 23:39:30 -0700195struct cgroup {
196 unsigned long flags; /* "unsigned long" so bitops work */
197
Paul Menaged20a3902009-04-02 16:57:22 -0700198 /*
199 * count users of this cgroup. >0 means busy, but doesn't
200 * necessarily indicate the number of tasks in the cgroup
201 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700202 atomic_t count;
203
204 /*
205 * We link our 'sibling' struct into our parent's 'children'.
206 * Our children link their 'sibling' into our 'children'.
207 */
208 struct list_head sibling; /* my parent's children */
209 struct list_head children; /* my children */
210
Paul Menaged20a3902009-04-02 16:57:22 -0700211 struct cgroup *parent; /* my parent */
Arnd Bergmann2c392b82010-02-24 19:41:39 +0100212 struct dentry __rcu *dentry; /* cgroup fs entry, RCU protected */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700213
214 /* Private pointers for each registered subsystem */
215 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
216
217 struct cgroupfs_root *root;
218 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700219
220 /*
221 * List of cg_cgroup_links pointing at css_sets with
222 * tasks in this cgroup. Protected by css_set_lock
223 */
224 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700225
226 /*
227 * Linked list running through all cgroups that can
228 * potentially be reaped by the release agent. Protected by
229 * release_list_lock
230 */
231 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700232
Ben Blum72a8cb32009-09-23 15:56:27 -0700233 /*
234 * list of pidlists, up to two for each namespace (one for procs, one
235 * for tasks); created on demand.
236 */
237 struct list_head pidlists;
238 struct mutex pidlist_mutex;
Paul Menagea47295e2009-01-07 18:07:44 -0800239
240 /* For RCU-protected deletion */
241 struct rcu_head rcu_head;
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800242
243 /* List of events which userspace want to recieve */
244 struct list_head event_list;
245 spinlock_t event_list_lock;
Paul Menage817929e2007-10-18 23:39:36 -0700246};
247
Paul Menaged20a3902009-04-02 16:57:22 -0700248/*
249 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700250 * cgroup_subsys_state objects. This saves space in the task struct
251 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700252 * list_add()/del() can bump the reference count on the entire cgroup
253 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700254 */
255
256struct css_set {
257
258 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700259 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700260
261 /*
Li Zefan472b1052008-04-29 01:00:11 -0700262 * List running through all cgroup groups in the same hash
263 * slot. Protected by css_set_lock
264 */
265 struct hlist_node hlist;
266
267 /*
Paul Menage817929e2007-10-18 23:39:36 -0700268 * List running through all tasks using this cgroup
269 * group. Protected by css_set_lock
270 */
271 struct list_head tasks;
272
273 /*
274 * List of cg_cgroup_link objects on link chains from
275 * cgroups referenced from this css_set. Protected by
276 * css_set_lock
277 */
278 struct list_head cg_links;
279
280 /*
281 * Set of subsystem states, one for each subsystem. This array
282 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800283 * during subsystem registration (at boot time) and modular subsystem
284 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700285 */
286 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700287
288 /* For RCU-protected deletion */
289 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700290};
291
Paul Menage91796562008-04-29 01:00:01 -0700292/*
293 * cgroup_map_cb is an abstract callback API for reporting map-valued
294 * control files
295 */
296
297struct cgroup_map_cb {
298 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
299 void *state;
300};
301
Paul Menaged20a3902009-04-02 16:57:22 -0700302/*
303 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700304 *
305 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800306 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700307 * - the 'cftype' of the file is file->f_dentry->d_fsdata
308 */
309
310#define MAX_CFTYPE_NAME 64
311struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700312 /*
313 * By convention, the name should begin with the name of the
314 * subsystem, followed by a period
315 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700316 char name[MAX_CFTYPE_NAME];
317 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700318 /*
319 * If not 0, file mode is set to this value, otherwise it will
320 * be figured out automatically
321 */
322 mode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700323
324 /*
325 * If non-zero, defines the maximum length of string that can
326 * be passed to write_string; defaults to 64
327 */
328 size_t max_write_len;
329
Paul Menagece16b492008-07-25 01:46:57 -0700330 int (*open)(struct inode *inode, struct file *file);
331 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
332 struct file *file,
333 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700334 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700335 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700336 * single integer. Use it in place of read()
337 */
Paul Menagece16b492008-07-25 01:46:57 -0700338 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700339 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700340 * read_s64() is a signed version of read_u64()
341 */
Paul Menagece16b492008-07-25 01:46:57 -0700342 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700343 /*
Paul Menage91796562008-04-29 01:00:01 -0700344 * read_map() is used for defining a map of key/value
345 * pairs. It should call cb->fill(cb, key, value) for each
346 * entry. The key/value pairs (and their ordering) should not
347 * change between reboots.
348 */
Paul Menagece16b492008-07-25 01:46:57 -0700349 int (*read_map)(struct cgroup *cont, struct cftype *cft,
350 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700351 /*
352 * read_seq_string() is used for outputting a simple sequence
353 * using seqfile.
354 */
Paul Menagece16b492008-07-25 01:46:57 -0700355 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
356 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700357
Paul Menagece16b492008-07-25 01:46:57 -0700358 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
359 struct file *file,
360 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700361
362 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700363 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700364 * a single integer (as parsed by simple_strtoull) from
365 * userspace. Use in place of write(); return 0 or error.
366 */
Paul Menagece16b492008-07-25 01:46:57 -0700367 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700368 /*
369 * write_s64() is a signed version of write_u64()
370 */
Paul Menagece16b492008-07-25 01:46:57 -0700371 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700372
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700373 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700374 * write_string() is passed a nul-terminated kernelspace
375 * buffer of maximum length determined by max_write_len.
376 * Returns 0 or -ve error code.
377 */
378 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
379 const char *buffer);
380 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700381 * trigger() callback can be used to get some kick from the
382 * userspace, when the actual string written is not important
383 * at all. The private field can be used to determine the
384 * kick type for multiplexing.
385 */
386 int (*trigger)(struct cgroup *cgrp, unsigned int event);
387
Paul Menagece16b492008-07-25 01:46:57 -0700388 int (*release)(struct inode *inode, struct file *file);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800389
390 /*
391 * register_event() callback will be used to add new userspace
392 * waiter for changes related to the cftype. Implement it if
393 * you want to provide this functionality. Use eventfd_signal()
394 * on eventfd to send notification to userspace.
395 */
396 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
397 struct eventfd_ctx *eventfd, const char *args);
398 /*
399 * unregister_event() callback will be called when userspace
400 * closes the eventfd or on cgroup removing.
401 * This callback must be implemented, if you want provide
402 * notification functionality.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800403 */
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700404 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800405 struct eventfd_ctx *eventfd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700406};
407
Cliff Wickman31a7df02008-02-07 00:14:42 -0800408struct cgroup_scanner {
409 struct cgroup *cg;
410 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
411 void (*process_task)(struct task_struct *p,
412 struct cgroup_scanner *scan);
413 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700414 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800415};
416
Paul Menaged20a3902009-04-02 16:57:22 -0700417/*
418 * Add a new file to the given cgroup directory. Should only be
419 * called by subsystems from within a populate() method
420 */
Li Zefanffd2d882008-02-23 15:24:09 -0800421int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700422 const struct cftype *cft);
423
Paul Menaged20a3902009-04-02 16:57:22 -0700424/*
425 * Add a set of new files to the given cgroup directory. Should
426 * only be called by subsystems from within a populate() method
427 */
Li Zefanffd2d882008-02-23 15:24:09 -0800428int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700429 struct cgroup_subsys *subsys,
430 const struct cftype cft[],
431 int count);
432
Li Zefanffd2d882008-02-23 15:24:09 -0800433int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700434
Li Zefanffd2d882008-02-23 15:24:09 -0800435int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700436
Li Zefanffd2d882008-02-23 15:24:09 -0800437int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700438
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700439/* Return true if cgrp is a descendant of the task's cgroup */
440int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700441
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100442/*
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700443 * When the subsys has to access css and may add permanent refcnt to css,
444 * it should take care of racy conditions with rmdir(). Following set of
445 * functions, is for stop/restart rmdir if necessary.
446 * Because these will call css_get/put, "css" should be alive css.
447 *
448 * cgroup_exclude_rmdir();
449 * ...do some jobs which may access arbitrary empty cgroup
450 * cgroup_release_and_wakeup_rmdir();
451 *
452 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
453 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
454 */
455
456void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
457void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
458
459/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100460 * Control Group subsystem type.
461 * See Documentation/cgroups/cgroups.txt for details
462 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700463
464struct cgroup_subsys {
465 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800466 struct cgroup *cgrp);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700467 int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Li Zefanffd2d882008-02-23 15:24:09 -0800468 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Ben Blumbe367d02009-09-23 15:56:31 -0700469 int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
470 struct task_struct *tsk, bool threadgroup);
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800471 void (*cancel_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
472 struct task_struct *tsk, bool threadgroup);
Li Zefanffd2d882008-02-23 15:24:09 -0800473 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -0700474 struct cgroup *old_cgrp, struct task_struct *tsk,
475 bool threadgroup);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700476 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
477 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
478 int (*populate)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800479 struct cgroup *cgrp);
480 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700481 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
Hugh Dickinse5991372009-01-06 14:39:22 -0800482
Paul Menageddbcc7e2007-10-18 23:39:30 -0700483 int subsys_id;
484 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700485 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700486 int early_init;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700487 /*
488 * True if this subsys uses ID. ID is not available before cgroup_init()
489 * (not available in early_init time.)
490 */
491 bool use_id;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700492#define MAX_CGROUP_TYPE_NAMELEN 32
493 const char *name;
494
Paul Menage999cd8a2009-01-07 18:08:36 -0800495 /*
496 * Protects sibling/children links of cgroups in this
497 * hierarchy, plus protects which hierarchy (or none) the
498 * subsystem is a part of (i.e. root/sibling). To avoid
499 * potential deadlocks, the following operations should not be
500 * undertaken while holding any hierarchy_mutex:
501 *
502 * - allocating memory
503 * - initiating hotplug events
504 */
505 struct mutex hierarchy_mutex;
Li Zefancfebe562009-02-11 13:04:36 -0800506 struct lock_class_key subsys_key;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700507
Paul Menage999cd8a2009-01-07 18:08:36 -0800508 /*
509 * Link to parent, and list entry in parent's children.
510 * Protected by this->hierarchy_mutex and cgroup_lock()
511 */
512 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700513 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700514 /* used when use_id == true */
515 struct idr idr;
516 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800517
518 /* should be defined only by modular subsystems */
519 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700520};
521
522#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
523#include <linux/cgroup_subsys.h>
524#undef SUBSYS
525
526static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800527 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700528{
Li Zefanffd2d882008-02-23 15:24:09 -0800529 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700530}
531
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200532/*
533 * function to get the cgroup_subsys_state which allows for extra
534 * rcu_dereference_check() conditions, such as locks used during the
535 * cgroup_subsys::attach() methods.
536 */
537#define task_subsys_state_check(task, subsys_id, __c) \
538 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
539 rcu_read_lock_held() || \
540 lockdep_is_held(&task->alloc_lock) || \
541 cgroup_lock_is_held() || (__c))
542
543static inline struct cgroup_subsys_state *
544task_subsys_state(struct task_struct *task, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700545{
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200546 return task_subsys_state_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700547}
548
549static inline struct cgroup* task_cgroup(struct task_struct *task,
550 int subsys_id)
551{
552 return task_subsys_state(task, subsys_id)->cgroup;
553}
554
Serge E. Hallyne885dcd2008-07-25 01:47:06 -0700555int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
556 char *nodename);
Paul Menage697f4162007-10-18 23:39:34 -0700557
Paul Menage817929e2007-10-18 23:39:36 -0700558/* A cgroup_iter should be treated as an opaque object */
559struct cgroup_iter {
560 struct list_head *cg_link;
561 struct list_head *task;
562};
563
Paul Menaged20a3902009-04-02 16:57:22 -0700564/*
565 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700566 *
567 * 1) call cgroup_iter_start to intialize an iterator
568 *
569 * 2) call cgroup_iter_next() to retrieve member tasks until it
570 * returns NULL or until you want to end the iteration
571 *
572 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800573 *
Paul Menaged20a3902009-04-02 16:57:22 -0700574 * Or, call cgroup_scan_tasks() to iterate through every task in a
575 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
576 * the test_task() callback, but not while calling the process_task()
577 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700578 */
Li Zefanffd2d882008-02-23 15:24:09 -0800579void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
580struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700581 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800582void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800583int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800584int cgroup_attach_task(struct cgroup *, struct task_struct *);
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700585int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
586
587static inline int cgroup_attach_task_current_cg(struct task_struct *tsk)
588{
589 return cgroup_attach_task_all(current, tsk);
590}
Paul Menage817929e2007-10-18 23:39:36 -0700591
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700592/*
593 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
594 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
595 * CSS ID is assigned at cgroup allocation (create) automatically
596 * and removed when subsys calls free_css_id() function. This is because
597 * the lifetime of cgroup_subsys_state is subsys's matter.
598 *
599 * Looking up and scanning function should be called under rcu_read_lock().
600 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
601 * But the css returned by this routine can be "not populated yet" or "being
602 * destroyed". The caller should check css and cgroup's status.
603 */
604
605/*
606 * Typically Called at ->destroy(), or somewhere the subsys frees
607 * cgroup_subsys_state.
608 */
609void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
610
611/* Find a cgroup_subsys_state which has given ID */
612
613struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
614
615/*
616 * Get a cgroup whose id is greater than or equal to id under tree of root.
617 * Returning a cgroup_subsys_state or NULL.
618 */
619struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
620 struct cgroup_subsys_state *root, int *foundid);
621
622/* Returns true if root is ancestor of cg */
623bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700624 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700625
626/* Get id and depth of css */
627unsigned short css_id(struct cgroup_subsys_state *css);
628unsigned short css_depth(struct cgroup_subsys_state *css);
629
Paul Menageddbcc7e2007-10-18 23:39:30 -0700630#else /* !CONFIG_CGROUPS */
631
632static inline int cgroup_init_early(void) { return 0; }
633static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700634static inline void cgroup_fork(struct task_struct *p) {}
635static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700636static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700637static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700638
639static inline void cgroup_lock(void) {}
640static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700641static inline int cgroupstats_build(struct cgroupstats *stats,
642 struct dentry *dentry)
643{
644 return -EINVAL;
645}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700646
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200647/* No cgroups - nothing to do */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -0700648static inline int cgroup_attach_task_all(struct task_struct *from,
649 struct task_struct *t)
650{
651 return 0;
652}
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200653static inline int cgroup_attach_task_current_cg(struct task_struct *t)
654{
655 return 0;
656}
657
Paul Menageddbcc7e2007-10-18 23:39:30 -0700658#endif /* !CONFIG_CGROUPS */
659
Paul Menageddbcc7e2007-10-18 23:39:30 -0700660#endif /* _LINUX_CGROUP_H */