]>
Commit | Line | Data |
---|---|---|
ddbcc7e8 PM |
1 | #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> | |
ddbcc7e8 PM |
12 | #include <linux/cpumask.h> |
13 | #include <linux/nodemask.h> | |
14 | #include <linux/rcupdate.h> | |
846c7bb0 | 15 | #include <linux/cgroupstats.h> |
31a7df01 | 16 | #include <linux/prio_heap.h> |
cc31edce | 17 | #include <linux/rwsem.h> |
ddbcc7e8 PM |
18 | |
19 | #ifdef CONFIG_CGROUPS | |
20 | ||
21 | struct cgroupfs_root; | |
22 | struct cgroup_subsys; | |
23 | struct inode; | |
84eea842 | 24 | struct cgroup; |
ddbcc7e8 PM |
25 | |
26 | extern int cgroup_init_early(void); | |
27 | extern int cgroup_init(void); | |
ddbcc7e8 | 28 | extern void cgroup_lock(void); |
84eea842 | 29 | extern bool cgroup_lock_live_group(struct cgroup *cgrp); |
ddbcc7e8 | 30 | extern void cgroup_unlock(void); |
b4f48b63 PM |
31 | extern void cgroup_fork(struct task_struct *p); |
32 | extern void cgroup_fork_callbacks(struct task_struct *p); | |
817929ec | 33 | extern void cgroup_post_fork(struct task_struct *p); |
b4f48b63 | 34 | extern void cgroup_exit(struct task_struct *p, int run_callbacks); |
846c7bb0 BS |
35 | extern int cgroupstats_build(struct cgroupstats *stats, |
36 | struct dentry *dentry); | |
ddbcc7e8 | 37 | |
a424316c PM |
38 | extern struct file_operations proc_cgroup_operations; |
39 | ||
817929ec PM |
40 | /* Define the enumeration of all cgroup subsystems */ |
41 | #define SUBSYS(_x) _x ## _subsys_id, | |
42 | enum cgroup_subsys_id { | |
43 | #include <linux/cgroup_subsys.h> | |
44 | CGROUP_SUBSYS_COUNT | |
45 | }; | |
46 | #undef SUBSYS | |
47 | ||
ddbcc7e8 PM |
48 | /* Per-subsystem/per-cgroup state maintained by the system. */ |
49 | struct 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 */ | |
65 | enum { | |
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 | ||
74 | static 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 | ||
81a6a5cd | 85 | extern void __css_put(struct cgroup_subsys_state *css); |
ddbcc7e8 PM |
86 | static inline void css_put(struct cgroup_subsys_state *css) |
87 | { | |
88 | if (!test_bit(CSS_ROOT, &css->flags)) | |
81a6a5cd | 89 | __css_put(css); |
ddbcc7e8 PM |
90 | } |
91 | ||
3116f0e3 PM |
92 | /* bits in struct cgroup flags field */ |
93 | enum { | |
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 | ||
ddbcc7e8 PM |
103 | struct 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 */ | |
a47295e6 | 119 | struct dentry *dentry; /* cgroup fs entry, RCU protected */ |
ddbcc7e8 PM |
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; | |
817929ec PM |
126 | |
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; | |
81a6a5cd PM |
132 | |
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; | |
cc31edce PM |
139 | |
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; | |
a47295e6 PM |
148 | |
149 | /* For RCU-protected deletion */ | |
150 | struct rcu_head rcu_head; | |
817929ec PM |
151 | }; |
152 | ||
153 | /* A css_set is a structure holding pointers to a set of | |
154 | * cgroup_subsys_state objects. This saves space in the task struct | |
155 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
156 | * list_add()/del() can bump the reference count on the entire | |
157 | * cgroup set for a task. | |
158 | */ | |
159 | ||
160 | struct css_set { | |
161 | ||
162 | /* Reference count */ | |
146aa1bd | 163 | atomic_t refcount; |
817929ec | 164 | |
472b1053 LZ |
165 | /* |
166 | * List running through all cgroup groups in the same hash | |
167 | * slot. Protected by css_set_lock | |
168 | */ | |
169 | struct hlist_node hlist; | |
170 | ||
817929ec PM |
171 | /* |
172 | * List running through all tasks using this cgroup | |
173 | * group. Protected by css_set_lock | |
174 | */ | |
175 | struct list_head tasks; | |
176 | ||
177 | /* | |
178 | * List of cg_cgroup_link objects on link chains from | |
179 | * cgroups referenced from this css_set. Protected by | |
180 | * css_set_lock | |
181 | */ | |
182 | struct list_head cg_links; | |
183 | ||
184 | /* | |
185 | * Set of subsystem states, one for each subsystem. This array | |
186 | * is immutable after creation apart from the init_css_set | |
187 | * during subsystem registration (at boot time). | |
188 | */ | |
189 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; | |
ddbcc7e8 PM |
190 | }; |
191 | ||
91796569 PM |
192 | /* |
193 | * cgroup_map_cb is an abstract callback API for reporting map-valued | |
194 | * control files | |
195 | */ | |
196 | ||
197 | struct cgroup_map_cb { | |
198 | int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value); | |
199 | void *state; | |
200 | }; | |
201 | ||
ddbcc7e8 PM |
202 | /* struct cftype: |
203 | * | |
204 | * The files in the cgroup filesystem mostly have a very simple read/write | |
205 | * handling, some common function will take care of it. Nevertheless some cases | |
206 | * (read tasks) are special and therefore I define this structure for every | |
207 | * kind of file. | |
208 | * | |
209 | * | |
210 | * When reading/writing to a file: | |
a043e3b2 | 211 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
ddbcc7e8 PM |
212 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
213 | */ | |
214 | ||
215 | #define MAX_CFTYPE_NAME 64 | |
216 | struct cftype { | |
217 | /* By convention, the name should begin with the name of the | |
218 | * subsystem, followed by a period */ | |
219 | char name[MAX_CFTYPE_NAME]; | |
220 | int private; | |
db3b1497 PM |
221 | |
222 | /* | |
223 | * If non-zero, defines the maximum length of string that can | |
224 | * be passed to write_string; defaults to 64 | |
225 | */ | |
226 | size_t max_write_len; | |
227 | ||
ce16b49d PM |
228 | int (*open)(struct inode *inode, struct file *file); |
229 | ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft, | |
230 | struct file *file, | |
231 | char __user *buf, size_t nbytes, loff_t *ppos); | |
ddbcc7e8 | 232 | /* |
f4c753b7 | 233 | * read_u64() is a shortcut for the common case of returning a |
ddbcc7e8 PM |
234 | * single integer. Use it in place of read() |
235 | */ | |
ce16b49d | 236 | u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft); |
e73d2c61 PM |
237 | /* |
238 | * read_s64() is a signed version of read_u64() | |
239 | */ | |
ce16b49d | 240 | s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft); |
91796569 PM |
241 | /* |
242 | * read_map() is used for defining a map of key/value | |
243 | * pairs. It should call cb->fill(cb, key, value) for each | |
244 | * entry. The key/value pairs (and their ordering) should not | |
245 | * change between reboots. | |
246 | */ | |
ce16b49d PM |
247 | int (*read_map)(struct cgroup *cont, struct cftype *cft, |
248 | struct cgroup_map_cb *cb); | |
29486df3 SH |
249 | /* |
250 | * read_seq_string() is used for outputting a simple sequence | |
251 | * using seqfile. | |
252 | */ | |
ce16b49d PM |
253 | int (*read_seq_string)(struct cgroup *cont, struct cftype *cft, |
254 | struct seq_file *m); | |
91796569 | 255 | |
ce16b49d PM |
256 | ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft, |
257 | struct file *file, | |
258 | const char __user *buf, size_t nbytes, loff_t *ppos); | |
355e0c48 PM |
259 | |
260 | /* | |
f4c753b7 | 261 | * write_u64() is a shortcut for the common case of accepting |
355e0c48 PM |
262 | * a single integer (as parsed by simple_strtoull) from |
263 | * userspace. Use in place of write(); return 0 or error. | |
264 | */ | |
ce16b49d | 265 | int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val); |
e73d2c61 PM |
266 | /* |
267 | * write_s64() is a signed version of write_u64() | |
268 | */ | |
ce16b49d | 269 | int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val); |
355e0c48 | 270 | |
db3b1497 PM |
271 | /* |
272 | * write_string() is passed a nul-terminated kernelspace | |
273 | * buffer of maximum length determined by max_write_len. | |
274 | * Returns 0 or -ve error code. | |
275 | */ | |
276 | int (*write_string)(struct cgroup *cgrp, struct cftype *cft, | |
277 | const char *buffer); | |
d447ea2f PE |
278 | /* |
279 | * trigger() callback can be used to get some kick from the | |
280 | * userspace, when the actual string written is not important | |
281 | * at all. The private field can be used to determine the | |
282 | * kick type for multiplexing. | |
283 | */ | |
284 | int (*trigger)(struct cgroup *cgrp, unsigned int event); | |
285 | ||
ce16b49d | 286 | int (*release)(struct inode *inode, struct file *file); |
ddbcc7e8 PM |
287 | }; |
288 | ||
31a7df01 CW |
289 | struct cgroup_scanner { |
290 | struct cgroup *cg; | |
291 | int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan); | |
292 | void (*process_task)(struct task_struct *p, | |
293 | struct cgroup_scanner *scan); | |
294 | struct ptr_heap *heap; | |
295 | }; | |
296 | ||
ddbcc7e8 PM |
297 | /* Add a new file to the given cgroup directory. Should only be |
298 | * called by subsystems from within a populate() method */ | |
ffd2d883 | 299 | int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys, |
ddbcc7e8 PM |
300 | const struct cftype *cft); |
301 | ||
302 | /* Add a set of new files to the given cgroup directory. Should | |
303 | * only be called by subsystems from within a populate() method */ | |
ffd2d883 | 304 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
305 | struct cgroup_subsys *subsys, |
306 | const struct cftype cft[], | |
307 | int count); | |
308 | ||
ffd2d883 | 309 | int cgroup_is_removed(const struct cgroup *cgrp); |
ddbcc7e8 | 310 | |
ffd2d883 | 311 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen); |
ddbcc7e8 | 312 | |
ffd2d883 | 313 | int cgroup_task_count(const struct cgroup *cgrp); |
bbcb81d0 | 314 | |
ddbcc7e8 | 315 | /* Return true if the cgroup is a descendant of the current cgroup */ |
ffd2d883 | 316 | int cgroup_is_descendant(const struct cgroup *cgrp); |
ddbcc7e8 PM |
317 | |
318 | /* Control Group subsystem type. See Documentation/cgroups.txt for details */ | |
319 | ||
320 | struct cgroup_subsys { | |
321 | struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss, | |
ffd2d883 LZ |
322 | struct cgroup *cgrp); |
323 | void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp); | |
324 | void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp); | |
ddbcc7e8 | 325 | int (*can_attach)(struct cgroup_subsys *ss, |
ffd2d883 LZ |
326 | struct cgroup *cgrp, struct task_struct *tsk); |
327 | void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
328 | struct cgroup *old_cgrp, struct task_struct *tsk); | |
ddbcc7e8 PM |
329 | void (*fork)(struct cgroup_subsys *ss, struct task_struct *task); |
330 | void (*exit)(struct cgroup_subsys *ss, struct task_struct *task); | |
331 | int (*populate)(struct cgroup_subsys *ss, | |
ffd2d883 LZ |
332 | struct cgroup *cgrp); |
333 | void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp); | |
ddbcc7e8 | 334 | void (*bind)(struct cgroup_subsys *ss, struct cgroup *root); |
e5991371 | 335 | |
ddbcc7e8 PM |
336 | int subsys_id; |
337 | int active; | |
8bab8dde | 338 | int disabled; |
ddbcc7e8 PM |
339 | int early_init; |
340 | #define MAX_CGROUP_TYPE_NAMELEN 32 | |
341 | const char *name; | |
342 | ||
999cd8a4 PM |
343 | /* |
344 | * Protects sibling/children links of cgroups in this | |
345 | * hierarchy, plus protects which hierarchy (or none) the | |
346 | * subsystem is a part of (i.e. root/sibling). To avoid | |
347 | * potential deadlocks, the following operations should not be | |
348 | * undertaken while holding any hierarchy_mutex: | |
349 | * | |
350 | * - allocating memory | |
351 | * - initiating hotplug events | |
352 | */ | |
353 | struct mutex hierarchy_mutex; | |
ddbcc7e8 | 354 | |
999cd8a4 PM |
355 | /* |
356 | * Link to parent, and list entry in parent's children. | |
357 | * Protected by this->hierarchy_mutex and cgroup_lock() | |
358 | */ | |
359 | struct cgroupfs_root *root; | |
ddbcc7e8 | 360 | struct list_head sibling; |
ddbcc7e8 PM |
361 | }; |
362 | ||
363 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys; | |
364 | #include <linux/cgroup_subsys.h> | |
365 | #undef SUBSYS | |
366 | ||
367 | static inline struct cgroup_subsys_state *cgroup_subsys_state( | |
ffd2d883 | 368 | struct cgroup *cgrp, int subsys_id) |
ddbcc7e8 | 369 | { |
ffd2d883 | 370 | return cgrp->subsys[subsys_id]; |
ddbcc7e8 PM |
371 | } |
372 | ||
373 | static inline struct cgroup_subsys_state *task_subsys_state( | |
374 | struct task_struct *task, int subsys_id) | |
375 | { | |
817929ec | 376 | return rcu_dereference(task->cgroups->subsys[subsys_id]); |
ddbcc7e8 PM |
377 | } |
378 | ||
379 | static inline struct cgroup* task_cgroup(struct task_struct *task, | |
380 | int subsys_id) | |
381 | { | |
382 | return task_subsys_state(task, subsys_id)->cgroup; | |
383 | } | |
384 | ||
e885dcde SH |
385 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss, |
386 | char *nodename); | |
697f4161 | 387 | |
817929ec PM |
388 | /* A cgroup_iter should be treated as an opaque object */ |
389 | struct cgroup_iter { | |
390 | struct list_head *cg_link; | |
391 | struct list_head *task; | |
392 | }; | |
393 | ||
394 | /* To iterate across the tasks in a cgroup: | |
395 | * | |
396 | * 1) call cgroup_iter_start to intialize an iterator | |
397 | * | |
398 | * 2) call cgroup_iter_next() to retrieve member tasks until it | |
399 | * returns NULL or until you want to end the iteration | |
400 | * | |
401 | * 3) call cgroup_iter_end() to destroy the iterator. | |
31a7df01 CW |
402 | * |
403 | * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset. | |
404 | * - cgroup_scan_tasks() holds the css_set_lock when calling the test_task() | |
405 | * callback, but not while calling the process_task() callback. | |
817929ec | 406 | */ |
ffd2d883 LZ |
407 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it); |
408 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, | |
817929ec | 409 | struct cgroup_iter *it); |
ffd2d883 | 410 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it); |
31a7df01 | 411 | int cgroup_scan_tasks(struct cgroup_scanner *scan); |
956db3ca | 412 | int cgroup_attach_task(struct cgroup *, struct task_struct *); |
817929ec | 413 | |
ddbcc7e8 PM |
414 | #else /* !CONFIG_CGROUPS */ |
415 | ||
416 | static inline int cgroup_init_early(void) { return 0; } | |
417 | static inline int cgroup_init(void) { return 0; } | |
b4f48b63 PM |
418 | static inline void cgroup_fork(struct task_struct *p) {} |
419 | static inline void cgroup_fork_callbacks(struct task_struct *p) {} | |
817929ec | 420 | static inline void cgroup_post_fork(struct task_struct *p) {} |
b4f48b63 | 421 | static inline void cgroup_exit(struct task_struct *p, int callbacks) {} |
ddbcc7e8 PM |
422 | |
423 | static inline void cgroup_lock(void) {} | |
424 | static inline void cgroup_unlock(void) {} | |
846c7bb0 BS |
425 | static inline int cgroupstats_build(struct cgroupstats *stats, |
426 | struct dentry *dentry) | |
427 | { | |
428 | return -EINVAL; | |
429 | } | |
ddbcc7e8 PM |
430 | |
431 | #endif /* !CONFIG_CGROUPS */ | |
432 | ||
433 | #endif /* _LINUX_CGROUP_H */ |