blob: 1164963c3a85f5213edddc91b7def976cd4c8eab [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>
Paul Menageddbcc7e2007-10-18 23:39:30 -070018
19#ifdef CONFIG_CGROUPS
20
21struct cgroupfs_root;
22struct cgroup_subsys;
23struct inode;
Paul Menage84eea842008-07-25 01:47:00 -070024struct cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -070025
26extern int cgroup_init_early(void);
27extern int cgroup_init(void);
Paul Menageddbcc7e2007-10-18 23:39:30 -070028extern void cgroup_lock(void);
Paul Menage84eea842008-07-25 01:47:00 -070029extern bool cgroup_lock_live_group(struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -070030extern void cgroup_unlock(void);
Paul Menageb4f48b62007-10-18 23:39:33 -070031extern void cgroup_fork(struct task_struct *p);
32extern void cgroup_fork_callbacks(struct task_struct *p);
Paul Menage817929e2007-10-18 23:39:36 -070033extern void cgroup_post_fork(struct task_struct *p);
Paul Menageb4f48b62007-10-18 23:39:33 -070034extern void cgroup_exit(struct task_struct *p, int run_callbacks);
Balbir Singh846c7bb2007-10-18 23:39:44 -070035extern int cgroupstats_build(struct cgroupstats *stats,
36 struct dentry *dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -070037
Paul Menagea4243162007-10-18 23:39:35 -070038extern struct file_operations proc_cgroup_operations;
39
Paul Menage817929e2007-10-18 23:39:36 -070040/* Define the enumeration of all cgroup subsystems */
41#define SUBSYS(_x) _x ## _subsys_id,
42enum cgroup_subsys_id {
43#include <linux/cgroup_subsys.h>
44 CGROUP_SUBSYS_COUNT
45};
46#undef SUBSYS
47
Paul Menageddbcc7e2007-10-18 23:39:30 -070048/* Per-subsystem/per-cgroup state maintained by the system. */
49struct cgroup_subsys_state {
50 /* The cgroup that this subsystem is attached to. Useful
51 * for subsystems that want to know about the cgroup
52 * hierarchy structure */
53 struct cgroup *cgroup;
54
55 /* State maintained by the cgroup system to allow
56 * subsystems to be "busy". Should be accessed via css_get()
57 * and css_put() */
58
59 atomic_t refcnt;
60
61 unsigned long flags;
62};
63
64/* bits in struct cgroup_subsys_state flags field */
65enum {
66 CSS_ROOT, /* This CSS is the root of the subsystem */
67};
68
69/*
70 * Call css_get() to hold a reference on the cgroup;
71 *
72 */
73
74static inline void css_get(struct cgroup_subsys_state *css)
75{
76 /* We don't need to reference count the root state */
77 if (!test_bit(CSS_ROOT, &css->flags))
78 atomic_inc(&css->refcnt);
79}
80/*
81 * css_put() should be called to release a reference taken by
82 * css_get()
83 */
84
Paul Menage81a6a5c2007-10-18 23:39:38 -070085extern void __css_put(struct cgroup_subsys_state *css);
Paul Menageddbcc7e2007-10-18 23:39:30 -070086static inline void css_put(struct cgroup_subsys_state *css)
87{
88 if (!test_bit(CSS_ROOT, &css->flags))
Paul Menage81a6a5c2007-10-18 23:39:38 -070089 __css_put(css);
Paul Menageddbcc7e2007-10-18 23:39:30 -070090}
91
Paul Menage3116f0e2008-04-29 01:00:04 -070092/* bits in struct cgroup flags field */
93enum {
94 /* Control Group is dead */
95 CGRP_REMOVED,
96 /* Control Group has previously had a child cgroup or a task,
97 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
98 CGRP_RELEASABLE,
99 /* Control Group requires release notifications to userspace */
100 CGRP_NOTIFY_ON_RELEASE,
101};
102
Paul Menageddbcc7e2007-10-18 23:39:30 -0700103struct cgroup {
104 unsigned long flags; /* "unsigned long" so bitops work */
105
106 /* count users of this cgroup. >0 means busy, but doesn't
107 * necessarily indicate the number of tasks in the
108 * cgroup */
109 atomic_t count;
110
111 /*
112 * We link our 'sibling' struct into our parent's 'children'.
113 * Our children link their 'sibling' into our 'children'.
114 */
115 struct list_head sibling; /* my parent's children */
116 struct list_head children; /* my children */
117
118 struct cgroup *parent; /* my parent */
119 struct dentry *dentry; /* cgroup fs entry */
120
121 /* Private pointers for each registered subsystem */
122 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
123
124 struct cgroupfs_root *root;
125 struct cgroup *top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700126
127 /*
128 * List of cg_cgroup_links pointing at css_sets with
129 * tasks in this cgroup. Protected by css_set_lock
130 */
131 struct list_head css_sets;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700132
133 /*
134 * Linked list running through all cgroups that can
135 * potentially be reaped by the release agent. Protected by
136 * release_list_lock
137 */
138 struct list_head release_list;
Paul Menagecc31edc2008-10-18 20:28:04 -0700139
140 /* pids_mutex protects the fields below */
141 struct rw_semaphore pids_mutex;
142 /* Array of process ids in the cgroup */
143 pid_t *tasks_pids;
144 /* How many files are using the current tasks_pids array */
145 int pids_use_count;
146 /* Length of the current tasks_pids array */
147 int pids_length;
Paul Menage817929e2007-10-18 23:39:36 -0700148};
149
150/* A css_set is a structure holding pointers to a set of
151 * cgroup_subsys_state objects. This saves space in the task struct
152 * object and speeds up fork()/exit(), since a single inc/dec and a
153 * list_add()/del() can bump the reference count on the entire
154 * cgroup set for a task.
155 */
156
157struct css_set {
158
159 /* Reference count */
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700160 atomic_t refcount;
Paul Menage817929e2007-10-18 23:39:36 -0700161
162 /*
Li Zefan472b1052008-04-29 01:00:11 -0700163 * List running through all cgroup groups in the same hash
164 * slot. Protected by css_set_lock
165 */
166 struct hlist_node hlist;
167
168 /*
Paul Menage817929e2007-10-18 23:39:36 -0700169 * List running through all tasks using this cgroup
170 * group. Protected by css_set_lock
171 */
172 struct list_head tasks;
173
174 /*
175 * List of cg_cgroup_link objects on link chains from
176 * cgroups referenced from this css_set. Protected by
177 * css_set_lock
178 */
179 struct list_head cg_links;
180
181 /*
182 * Set of subsystem states, one for each subsystem. This array
183 * is immutable after creation apart from the init_css_set
184 * during subsystem registration (at boot time).
185 */
186 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700187};
188
Paul Menage91796562008-04-29 01:00:01 -0700189/*
190 * cgroup_map_cb is an abstract callback API for reporting map-valued
191 * control files
192 */
193
194struct cgroup_map_cb {
195 int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
196 void *state;
197};
198
Paul Menageddbcc7e2007-10-18 23:39:30 -0700199/* struct cftype:
200 *
201 * The files in the cgroup filesystem mostly have a very simple read/write
202 * handling, some common function will take care of it. Nevertheless some cases
203 * (read tasks) are special and therefore I define this structure for every
204 * kind of file.
205 *
206 *
207 * When reading/writing to a file:
Li Zefana043e3b2008-02-23 15:24:09 -0800208 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
Paul Menageddbcc7e2007-10-18 23:39:30 -0700209 * - the 'cftype' of the file is file->f_dentry->d_fsdata
210 */
211
212#define MAX_CFTYPE_NAME 64
213struct cftype {
214 /* By convention, the name should begin with the name of the
215 * subsystem, followed by a period */
216 char name[MAX_CFTYPE_NAME];
217 int private;
Paul Menagedb3b1492008-07-25 01:46:58 -0700218
219 /*
220 * If non-zero, defines the maximum length of string that can
221 * be passed to write_string; defaults to 64
222 */
223 size_t max_write_len;
224
Paul Menagece16b492008-07-25 01:46:57 -0700225 int (*open)(struct inode *inode, struct file *file);
226 ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
227 struct file *file,
228 char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700229 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700230 * read_u64() is a shortcut for the common case of returning a
Paul Menageddbcc7e2007-10-18 23:39:30 -0700231 * single integer. Use it in place of read()
232 */
Paul Menagece16b492008-07-25 01:46:57 -0700233 u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menage91796562008-04-29 01:00:01 -0700234 /*
Paul Menagee73d2c62008-04-29 01:00:06 -0700235 * read_s64() is a signed version of read_u64()
236 */
Paul Menagece16b492008-07-25 01:46:57 -0700237 s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
Paul Menagee73d2c62008-04-29 01:00:06 -0700238 /*
Paul Menage91796562008-04-29 01:00:01 -0700239 * read_map() is used for defining a map of key/value
240 * pairs. It should call cb->fill(cb, key, value) for each
241 * entry. The key/value pairs (and their ordering) should not
242 * change between reboots.
243 */
Paul Menagece16b492008-07-25 01:46:57 -0700244 int (*read_map)(struct cgroup *cont, struct cftype *cft,
245 struct cgroup_map_cb *cb);
Serge E. Hallyn29486df2008-04-29 01:00:14 -0700246 /*
247 * read_seq_string() is used for outputting a simple sequence
248 * using seqfile.
249 */
Paul Menagece16b492008-07-25 01:46:57 -0700250 int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
251 struct seq_file *m);
Paul Menage91796562008-04-29 01:00:01 -0700252
Paul Menagece16b492008-07-25 01:46:57 -0700253 ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
254 struct file *file,
255 const char __user *buf, size_t nbytes, loff_t *ppos);
Paul Menage355e0c42007-10-18 23:39:33 -0700256
257 /*
Paul Menagef4c753b2008-04-29 00:59:56 -0700258 * write_u64() is a shortcut for the common case of accepting
Paul Menage355e0c42007-10-18 23:39:33 -0700259 * a single integer (as parsed by simple_strtoull) from
260 * userspace. Use in place of write(); return 0 or error.
261 */
Paul Menagece16b492008-07-25 01:46:57 -0700262 int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
Paul Menagee73d2c62008-04-29 01:00:06 -0700263 /*
264 * write_s64() is a signed version of write_u64()
265 */
Paul Menagece16b492008-07-25 01:46:57 -0700266 int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
Paul Menage355e0c42007-10-18 23:39:33 -0700267
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700268 /*
Paul Menagedb3b1492008-07-25 01:46:58 -0700269 * write_string() is passed a nul-terminated kernelspace
270 * buffer of maximum length determined by max_write_len.
271 * Returns 0 or -ve error code.
272 */
273 int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
274 const char *buffer);
275 /*
Pavel Emelyanovd447ea22008-04-29 01:00:08 -0700276 * trigger() callback can be used to get some kick from the
277 * userspace, when the actual string written is not important
278 * at all. The private field can be used to determine the
279 * kick type for multiplexing.
280 */
281 int (*trigger)(struct cgroup *cgrp, unsigned int event);
282
Paul Menagece16b492008-07-25 01:46:57 -0700283 int (*release)(struct inode *inode, struct file *file);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700284};
285
Cliff Wickman31a7df02008-02-07 00:14:42 -0800286struct cgroup_scanner {
287 struct cgroup *cg;
288 int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
289 void (*process_task)(struct task_struct *p,
290 struct cgroup_scanner *scan);
291 struct ptr_heap *heap;
292};
293
Paul Menageddbcc7e2007-10-18 23:39:30 -0700294/* Add a new file to the given cgroup directory. Should only be
295 * called by subsystems from within a populate() method */
Li Zefanffd2d882008-02-23 15:24:09 -0800296int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700297 const struct cftype *cft);
298
299/* Add a set of new files to the given cgroup directory. Should
300 * only be called by subsystems from within a populate() method */
Li Zefanffd2d882008-02-23 15:24:09 -0800301int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -0700302 struct cgroup_subsys *subsys,
303 const struct cftype cft[],
304 int count);
305
Li Zefanffd2d882008-02-23 15:24:09 -0800306int cgroup_is_removed(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700307
Li Zefanffd2d882008-02-23 15:24:09 -0800308int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700309
Li Zefanffd2d882008-02-23 15:24:09 -0800310int cgroup_task_count(const struct cgroup *cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -0700311
Paul Menageddbcc7e2007-10-18 23:39:30 -0700312/* Return true if the cgroup is a descendant of the current cgroup */
Li Zefanffd2d882008-02-23 15:24:09 -0800313int cgroup_is_descendant(const struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700314
315/* Control Group subsystem type. See Documentation/cgroups.txt for details */
316
317struct cgroup_subsys {
318 struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800319 struct cgroup *cgrp);
320 void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
321 void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700322 int (*can_attach)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800323 struct cgroup *cgrp, struct task_struct *tsk);
324 void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
325 struct cgroup *old_cgrp, struct task_struct *tsk);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700326 void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
327 void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
328 int (*populate)(struct cgroup_subsys *ss,
Li Zefanffd2d882008-02-23 15:24:09 -0800329 struct cgroup *cgrp);
330 void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700331 void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700332 /*
333 * This routine is called with the task_lock of mm->owner held
334 */
335 void (*mm_owner_changed)(struct cgroup_subsys *ss,
336 struct cgroup *old,
Balbir Singh9363b9f2008-10-15 22:01:05 -0700337 struct cgroup *new,
338 struct task_struct *p);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700339 int subsys_id;
340 int active;
Paul Menage8bab8dd2008-04-04 14:29:57 -0700341 int disabled;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700342 int early_init;
343#define MAX_CGROUP_TYPE_NAMELEN 32
344 const char *name;
345
346 /* Protected by RCU */
347 struct cgroupfs_root *root;
348
349 struct list_head sibling;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700350};
351
352#define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
353#include <linux/cgroup_subsys.h>
354#undef SUBSYS
355
356static inline struct cgroup_subsys_state *cgroup_subsys_state(
Li Zefanffd2d882008-02-23 15:24:09 -0800357 struct cgroup *cgrp, int subsys_id)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700358{
Li Zefanffd2d882008-02-23 15:24:09 -0800359 return cgrp->subsys[subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700360}
361
362static inline struct cgroup_subsys_state *task_subsys_state(
363 struct task_struct *task, int subsys_id)
364{
Paul Menage817929e2007-10-18 23:39:36 -0700365 return rcu_dereference(task->cgroups->subsys[subsys_id]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700366}
367
368static inline struct cgroup* task_cgroup(struct task_struct *task,
369 int subsys_id)
370{
371 return task_subsys_state(task, subsys_id)->cgroup;
372}
373
Serge E. Hallyne885dcd2008-07-25 01:47:06 -0700374int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
375 char *nodename);
Paul Menage697f4162007-10-18 23:39:34 -0700376
Paul Menage817929e2007-10-18 23:39:36 -0700377/* A cgroup_iter should be treated as an opaque object */
378struct cgroup_iter {
379 struct list_head *cg_link;
380 struct list_head *task;
381};
382
383/* To iterate across the tasks in a cgroup:
384 *
385 * 1) call cgroup_iter_start to intialize an iterator
386 *
387 * 2) call cgroup_iter_next() to retrieve member tasks until it
388 * returns NULL or until you want to end the iteration
389 *
390 * 3) call cgroup_iter_end() to destroy the iterator.
Cliff Wickman31a7df02008-02-07 00:14:42 -0800391 *
392 * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
393 * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
394 * callback, but not while calling the process_task() callback.
Paul Menage817929e2007-10-18 23:39:36 -0700395 */
Li Zefanffd2d882008-02-23 15:24:09 -0800396void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
397struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700398 struct cgroup_iter *it);
Li Zefanffd2d882008-02-23 15:24:09 -0800399void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
Cliff Wickman31a7df02008-02-07 00:14:42 -0800400int cgroup_scan_tasks(struct cgroup_scanner *scan);
Cliff Wickman956db3c2008-02-07 00:14:43 -0800401int cgroup_attach_task(struct cgroup *, struct task_struct *);
Paul Menage817929e2007-10-18 23:39:36 -0700402
Paul Menage886465f2008-10-18 20:28:05 -0700403void cgroup_mm_owner_callbacks(struct task_struct *old,
404 struct task_struct *new);
405
Paul Menageddbcc7e2007-10-18 23:39:30 -0700406#else /* !CONFIG_CGROUPS */
407
408static inline int cgroup_init_early(void) { return 0; }
409static inline int cgroup_init(void) { return 0; }
Paul Menageb4f48b62007-10-18 23:39:33 -0700410static inline void cgroup_fork(struct task_struct *p) {}
411static inline void cgroup_fork_callbacks(struct task_struct *p) {}
Paul Menage817929e2007-10-18 23:39:36 -0700412static inline void cgroup_post_fork(struct task_struct *p) {}
Paul Menageb4f48b62007-10-18 23:39:33 -0700413static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700414
415static inline void cgroup_lock(void) {}
416static inline void cgroup_unlock(void) {}
Balbir Singh846c7bb2007-10-18 23:39:44 -0700417static inline int cgroupstats_build(struct cgroupstats *stats,
418 struct dentry *dentry)
419{
420 return -EINVAL;
421}
Paul Menageddbcc7e2007-10-18 23:39:30 -0700422
Paul Menage886465f2008-10-18 20:28:05 -0700423static inline void cgroup_mm_owner_callbacks(struct task_struct *old,
424 struct task_struct *new) {}
425
Paul Menageddbcc7e2007-10-18 23:39:30 -0700426#endif /* !CONFIG_CGROUPS */
427
Paul Menageddbcc7e2007-10-18 23:39:30 -0700428#endif /* _LINUX_CGROUP_H */