blob: ed3e92e41c6e5683ad3dbb823e48259f5150ac33 [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;
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800238
239 /* List of events which userspace want to recieve */
240 struct list_head event_list;
241 spinlock_t event_list_lock;
Paul Menage817929e2007-10-18 23:39:36 -0700242};
243
Paul Menaged20a3902009-04-02 16:57:22 -0700244/*
245 * A css_set is a structure holding pointers to a set of
Paul Menage817929e2007-10-18 23:39:36 -0700246 * cgroup_subsys_state objects. This saves space in the task struct
247 * object and speeds up fork()/exit(), since a single inc/dec and a
Paul Menaged20a3902009-04-02 16:57:22 -0700248 * list_add()/del() can bump the reference count on the entire cgroup
249 * set for a task.
Paul Menage817929e2007-10-18 23:39:36 -0700250 */
251
252struct css_set {
253
254 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700255 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700256
257 /*
Li Zefan472b1052008-04-29 01:00:11 -0700258 * List running through all cgroup groups in the same hash
259 * slot. Protected by css_set_lock
260 */
261 struct hlist_node hlist;
262
263 /*
Paul Menage817929e2007-10-18 23:39:36 -0700264 * List running through all tasks using this cgroup
265 * group. Protected by css_set_lock
266 */
267 struct list_head tasks;
268
269 /*
270 * List of cg_cgroup_link objects on link chains from
271 * cgroups referenced from this css_set. Protected by
272 * css_set_lock
273 */
274 struct list_head cg_links;
275
276 /*
277 * Set of subsystem states, one for each subsystem. This array
278 * is immutable after creation apart from the init_css_set
Ben Blumcf5d5942010-03-10 15:22:09 -0800279 * during subsystem registration (at boot time) and modular subsystem
280 * loading/unloading.
Paul Menage817929e2007-10-18 23:39:36 -0700281 */
282 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Ben Blumc3783692009-09-23 15:56:29 -0700283
284 /* For RCU-protected deletion */
285 struct rcu_head rcu_head;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700286};
287
Paul Menage91796562008-04-29 01:00:01 -0700288/*
289 * cgroup_map_cb is an abstract callback API for reporting map-valued
290 * control files
291 */
292
293struct cgroup_map_cb {
294 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
295 void *state;
296};
297
Paul Menaged20a3902009-04-02 16:57:22 -0700298/*
299 * struct cftype: handler definitions for cgroup control files
Paul Menageddbcc7e2007-10-18 23:39:30 -0700300 *
301 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800302 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700303 * - the 'cftype' of the file is file->f_dentry->d_fsdata
304 */
305
306#define MAX_CFTYPE_NAME 64
307struct cftype {
Paul Menaged20a3902009-04-02 16:57:22 -0700308 /*
309 * By convention, the name should begin with the name of the
310 * subsystem, followed by a period
311 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700312 char name[MAX_CFTYPE_NAME];
313 int private;
Li Zefan099fca32009-04-02 16:57:29 -0700314 /*
315 * If not 0, file mode is set to this value, otherwise it will
316 * be figured out automatically
317 */
318 mode_t mode;
Paul Menagedb3b1492008-07-25 01:46:58 -0700319
320 /*
321 * If non-zero, defines the maximum length of string that can
322 * be passed to write_string; defaults to 64
323 */
324 size_t max_write_len;
325
Paul Menagece16b492008-07-25 01:46:57 -0700326 int (*open)(struct inode *inode, struct file *file);
327 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
328 struct file *file,
329 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700330 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700331 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700332 * single integer. Use it in place of read()
333 */
Paul Menagece16b492008-07-25 01:46:57 -0700334 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700335 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700336 * read_s64() is a signed version of read_u64()
337 */
Paul Menagece16b492008-07-25 01:46:57 -0700338 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700339 /*
Paul Menage91796562008-04-29 01:00:01 -0700340 * read_map() is used for defining a map of key/value
341 * pairs. It should call cb->fill(cb, key, value) for each
342 * entry. The key/value pairs (and their ordering) should not
343 * change between reboots.
344 */
Paul Menagece16b492008-07-25 01:46:57 -0700345 int (*read_map)(struct cgroup *cont, struct cftype *cft,
346 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700347 /*
348 * read_seq_string() is used for outputting a simple sequence
349 * using seqfile.
350 */
Paul Menagece16b492008-07-25 01:46:57 -0700351 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
352 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700353
Paul Menagece16b492008-07-25 01:46:57 -0700354 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
355 struct file *file,
356 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700357
358 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700359 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700360 * a single integer (as parsed by simple_strtoull) from
361 * userspace. Use in place of write(); return 0 or error.
362 */
Paul Menagece16b492008-07-25 01:46:57 -0700363 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700364 /*
365 * write_s64() is a signed version of write_u64()
366 */
Paul Menagece16b492008-07-25 01:46:57 -0700367 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700368
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700369 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700370 * write_string() is passed a nul-terminated kernelspace
371 * buffer of maximum length determined by max_write_len.
372 * Returns 0 or -ve error code.
373 */
374 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
375 const char *buffer);
376 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700377 * trigger() callback can be used to get some kick from the
378 * userspace, when the actual string written is not important
379 * at all. The private field can be used to determine the
380 * kick type for multiplexing.
381 */
382 int (*trigger)(struct cgroup *cgrp, unsigned int event);
383
Paul Menagece16b492008-07-25 01:46:57 -0700384 int (*release)(struct inode *inode, struct file *file);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800385
386 /*
387 * register_event() callback will be used to add new userspace
388 * waiter for changes related to the cftype. Implement it if
389 * you want to provide this functionality. Use eventfd_signal()
390 * on eventfd to send notification to userspace.
391 */
392 int (*register_event)(struct cgroup *cgrp, struct cftype *cft,
393 struct eventfd_ctx *eventfd, const char *args);
394 /*
395 * unregister_event() callback will be called when userspace
396 * closes the eventfd or on cgroup removing.
397 * This callback must be implemented, if you want provide
398 * notification functionality.
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800399 */
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700400 void (*unregister_event)(struct cgroup *cgrp, struct cftype *cft,
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800401 struct eventfd_ctx *eventfd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700402};
403
Cliff Wickman31a7df02008-02-07 00:14:42 -0800404struct cgroup_scanner {
405 struct cgroup *cg;
406 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
407 void (*process_task)(struct task_struct *p,
408 struct cgroup_scanner *scan);
409 struct ptr_heap *heap;
Li Zefanbd1a8ab2009-04-02 16:57:50 -0700410 void *data;
Cliff Wickman31a7df02008-02-07 00:14:42 -0800411};
412
Paul Menaged20a3902009-04-02 16:57:22 -0700413/*
414 * Add a new file to the given cgroup directory. Should only be
415 * called by subsystems from within a populate() method
416 */
Li Zefanffd2d882008-02-23 15:24:09 -0800417int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700418 const struct cftype *cft);
419
Paul Menaged20a3902009-04-02 16:57:22 -0700420/*
421 * Add a set of new files to the given cgroup directory. Should
422 * only be called by subsystems from within a populate() method
423 */
Li Zefanffd2d882008-02-23 15:24:09 -0800424int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700425 struct cgroup_subsys *subsys,
426 const struct cftype cft[],
427 int count);
428
Li Zefanffd2d882008-02-23 15:24:09 -0800429int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700430
Li Zefanffd2d882008-02-23 15:24:09 -0800431int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700432
Li Zefanffd2d882008-02-23 15:24:09 -0800433int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700434
Grzegorz Nosek313e9242009-04-02 16:57:23 -0700435/* Return true if cgrp is a descendant of the task's cgroup */
436int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700437
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100438/*
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700439 * When the subsys has to access css and may add permanent refcnt to css,
440 * it should take care of racy conditions with rmdir(). Following set of
441 * functions, is for stop/restart rmdir if necessary.
442 * Because these will call css_get/put, "css" should be alive css.
443 *
444 * cgroup_exclude_rmdir();
445 * ...do some jobs which may access arbitrary empty cgroup
446 * cgroup_release_and_wakeup_rmdir();
447 *
448 * When someone removes a cgroup while cgroup_exclude_rmdir() holds it,
449 * it sleeps and cgroup_release_and_wakeup_rmdir() will wake him up.
450 */
451
452void cgroup_exclude_rmdir(struct cgroup_subsys_state *css);
453void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css);
454
455/*
Thadeu Lima de Souza Cascardo21acb9c2009-02-04 10:12:08 +0100456 * Control Group subsystem type.
457 * See Documentation/cgroups/cgroups.txt for details
458 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700459
460struct cgroup_subsys {
461 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800462 struct cgroup *cgrp);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700463 int (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Li Zefanffd2d882008-02-23 15:24:09 -0800464 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Ben Blumbe367d02009-09-23 15:56:31 -0700465 int (*can_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
466 struct task_struct *tsk, bool threadgroup);
Daisuke Nishimura2468c722010-03-10 15:22:03 -0800467 void (*cancel_attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
468 struct task_struct *tsk, bool threadgroup);
Li Zefanffd2d882008-02-23 15:24:09 -0800469 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
Ben Blumbe367d02009-09-23 15:56:31 -0700470 struct cgroup *old_cgrp, struct task_struct *tsk,
471 bool threadgroup);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700472 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
473 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
474 int (*populate)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800475 struct cgroup *cgrp);
476 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700477 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
Hugh Dickinse5991372009-01-06 14:39:22 -0800478
Paul Menageddbcc7e2007-10-18 23:39:30 -0700479 int subsys_id;
480 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700481 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700482 int early_init;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700483 /*
484 * True if this subsys uses ID. ID is not available before cgroup_init()
485 * (not available in early_init time.)
486 */
487 bool use_id;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700488#define MAX_CGROUP_TYPE_NAMELEN 32
489 const char *name;
490
Paul Menage999cd8a2009-01-07 18:08:36 -0800491 /*
492 * Protects sibling/children links of cgroups in this
493 * hierarchy, plus protects which hierarchy (or none) the
494 * subsystem is a part of (i.e. root/sibling). To avoid
495 * potential deadlocks, the following operations should not be
496 * undertaken while holding any hierarchy_mutex:
497 *
498 * - allocating memory
499 * - initiating hotplug events
500 */
501 struct mutex hierarchy_mutex;
Li Zefancfebe562009-02-11 13:04:36 -0800502 struct lock_class_key subsys_key;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700503
Paul Menage999cd8a2009-01-07 18:08:36 -0800504 /*
505 * Link to parent, and list entry in parent's children.
506 * Protected by this->hierarchy_mutex and cgroup_lock()
507 */
508 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700509 struct list_head sibling;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700510 /* used when use_id == true */
511 struct idr idr;
512 spinlock_t id_lock;
Ben Blume6a11052010-03-10 15:22:09 -0800513
514 /* should be defined only by modular subsystems */
515 struct module *module;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700516};
517
518#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
519#include <linux/cgroup_subsys.h>
520#undef SUBSYS
521
522static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800523 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700524{
Li Zefanffd2d882008-02-23 15:24:09 -0800525 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700526}
527
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200528/*
529 * function to get the cgroup_subsys_state which allows for extra
530 * rcu_dereference_check() conditions, such as locks used during the
531 * cgroup_subsys::attach() methods.
532 */
533#define task_subsys_state_check(task, subsys_id, __c) \
534 rcu_dereference_check(task->cgroups->subsys[subsys_id], \
535 rcu_read_lock_held() || \
536 lockdep_is_held(&task->alloc_lock) || \
537 cgroup_lock_is_held() || (__c))
538
539static inline struct cgroup_subsys_state *
540task_subsys_state(struct task_struct *task, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700541{
Peter Zijlstradc61b1d2010-06-08 11:40:42 +0200542 return task_subsys_state_check(task, subsys_id, false);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700543}
544
545static inline struct cgroup* task_cgroup(struct task_struct *task,
546 int subsys_id)
547{
548 return task_subsys_state(task, subsys_id)->cgroup;
549}
550
Serge E. Hallyne885dcd2008-07-25 01:47:06 -0700551int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
552 char *nodename);
Paul Menage697f4162007-10-18 23:39:34 -0700553
Paul Menage817929e2007-10-18 23:39:36 -0700554/* A cgroup_iter should be treated as an opaque object */
555struct cgroup_iter {
556 struct list_head *cg_link;
557 struct list_head *task;
558};
559
Paul Menaged20a3902009-04-02 16:57:22 -0700560/*
561 * To iterate across the tasks in a cgroup:
Paul Menage817929e2007-10-18 23:39:36 -0700562 *
563 * 1) call cgroup_iter_start to intialize an iterator
564 *
565 * 2) call cgroup_iter_next() to retrieve member tasks until it
566 * returns NULL or until you want to end the iteration
567 *
568 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800569 *
Paul Menaged20a3902009-04-02 16:57:22 -0700570 * Or, call cgroup_scan_tasks() to iterate through every task in a
571 * cgroup - cgroup_scan_tasks() holds the css_set_lock when calling
572 * the test_task() callback, but not while calling the process_task()
573 * callback.
Paul Menage817929e2007-10-18 23:39:36 -0700574 */
Li Zefanffd2d882008-02-23 15:24:09 -0800575void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
576struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700577 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800578void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800579int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800580int cgroup_attach_task(struct cgroup *, struct task_struct *);
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200581int cgroup_attach_task_current_cg(struct task_struct *);
Paul Menage817929e2007-10-18 23:39:36 -0700582
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700583/*
584 * CSS ID is ID for cgroup_subsys_state structs under subsys. This only works
585 * if cgroup_subsys.use_id == true. It can be used for looking up and scanning.
586 * CSS ID is assigned at cgroup allocation (create) automatically
587 * and removed when subsys calls free_css_id() function. This is because
588 * the lifetime of cgroup_subsys_state is subsys's matter.
589 *
590 * Looking up and scanning function should be called under rcu_read_lock().
591 * Taking cgroup_mutex()/hierarchy_mutex() is not necessary for following calls.
592 * But the css returned by this routine can be "not populated yet" or "being
593 * destroyed". The caller should check css and cgroup's status.
594 */
595
596/*
597 * Typically Called at ->destroy(), or somewhere the subsys frees
598 * cgroup_subsys_state.
599 */
600void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css);
601
602/* Find a cgroup_subsys_state which has given ID */
603
604struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id);
605
606/*
607 * Get a cgroup whose id is greater than or equal to id under tree of root.
608 * Returning a cgroup_subsys_state or NULL.
609 */
610struct cgroup_subsys_state *css_get_next(struct cgroup_subsys *ss, int id,
611 struct cgroup_subsys_state *root, int *foundid);
612
613/* Returns true if root is ancestor of cg */
614bool css_is_ancestor(struct cgroup_subsys_state *cg,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700615 const struct cgroup_subsys_state *root);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700616
617/* Get id and depth of css */
618unsigned short css_id(struct cgroup_subsys_state *css);
619unsigned short css_depth(struct cgroup_subsys_state *css);
620
Paul Menageddbcc7e2007-10-18 23:39:30 -0700621#else /* !CONFIG_CGROUPS */
622
623static inline int cgroup_init_early(void) { return 0; }
624static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700625static inline void cgroup_fork(struct task_struct *p) {}
626static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700627static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700628static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700629
630static inline void cgroup_lock(void) {}
631static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700632static inline int cgroupstats_build(struct cgroupstats *stats,
633 struct dentry *dentry)
634{
635 return -EINVAL;
636}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700637
Sridhar Samudralad7926ee2010-05-30 22:24:39 +0200638/* No cgroups - nothing to do */
639static inline int cgroup_attach_task_current_cg(struct task_struct *t)
640{
641 return 0;
642}
643
Paul Menageddbcc7e2007-10-18 23:39:30 -0700644#endif /* !CONFIG_CGROUPS */
645
Paul Menageddbcc7e2007-10-18 23:39:30 -0700646#endif /* _LINUX_CGROUP_H */