xref: /linux-6.15/include/linux/cgroup.h (revision d289bf7b)
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>
12 #include <linux/kref.h>
13 #include <linux/cpumask.h>
14 #include <linux/nodemask.h>
15 #include <linux/rcupdate.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/prio_heap.h>
18 
19 #ifdef CONFIG_CGROUPS
20 
21 struct cgroupfs_root;
22 struct cgroup_subsys;
23 struct inode;
24 
25 extern int cgroup_init_early(void);
26 extern int cgroup_init(void);
27 extern void cgroup_init_smp(void);
28 extern void cgroup_lock(void);
29 extern void cgroup_unlock(void);
30 extern void cgroup_fork(struct task_struct *p);
31 extern void cgroup_fork_callbacks(struct task_struct *p);
32 extern void cgroup_post_fork(struct task_struct *p);
33 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
34 extern int cgroupstats_build(struct cgroupstats *stats,
35 				struct dentry *dentry);
36 
37 extern struct file_operations proc_cgroup_operations;
38 
39 /* Define the enumeration of all cgroup subsystems */
40 #define SUBSYS(_x) _x ## _subsys_id,
41 enum cgroup_subsys_id {
42 #include <linux/cgroup_subsys.h>
43 	CGROUP_SUBSYS_COUNT
44 };
45 #undef SUBSYS
46 
47 /* Per-subsystem/per-cgroup state maintained by the system. */
48 struct cgroup_subsys_state {
49 	/* The cgroup that this subsystem is attached to. Useful
50 	 * for subsystems that want to know about the cgroup
51 	 * hierarchy structure */
52 	struct cgroup *cgroup;
53 
54 	/* State maintained by the cgroup system to allow
55 	 * subsystems to be "busy". Should be accessed via css_get()
56 	 * and css_put() */
57 
58 	atomic_t refcnt;
59 
60 	unsigned long flags;
61 };
62 
63 /* bits in struct cgroup_subsys_state flags field */
64 enum {
65 	CSS_ROOT, /* This CSS is the root of the subsystem */
66 };
67 
68 /*
69  * Call css_get() to hold a reference on the cgroup;
70  *
71  */
72 
73 static inline void css_get(struct cgroup_subsys_state *css)
74 {
75 	/* We don't need to reference count the root state */
76 	if (!test_bit(CSS_ROOT, &css->flags))
77 		atomic_inc(&css->refcnt);
78 }
79 /*
80  * css_put() should be called to release a reference taken by
81  * css_get()
82  */
83 
84 extern void __css_put(struct cgroup_subsys_state *css);
85 static inline void css_put(struct cgroup_subsys_state *css)
86 {
87 	if (!test_bit(CSS_ROOT, &css->flags))
88 		__css_put(css);
89 }
90 
91 struct cgroup {
92 	unsigned long flags;		/* "unsigned long" so bitops work */
93 
94 	/* count users of this cgroup. >0 means busy, but doesn't
95 	 * necessarily indicate the number of tasks in the
96 	 * cgroup */
97 	atomic_t count;
98 
99 	/*
100 	 * We link our 'sibling' struct into our parent's 'children'.
101 	 * Our children link their 'sibling' into our 'children'.
102 	 */
103 	struct list_head sibling;	/* my parent's children */
104 	struct list_head children;	/* my children */
105 
106 	struct cgroup *parent;	/* my parent */
107 	struct dentry *dentry;	  	/* cgroup fs entry */
108 
109 	/* Private pointers for each registered subsystem */
110 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
111 
112 	struct cgroupfs_root *root;
113 	struct cgroup *top_cgroup;
114 
115 	/*
116 	 * List of cg_cgroup_links pointing at css_sets with
117 	 * tasks in this cgroup. Protected by css_set_lock
118 	 */
119 	struct list_head css_sets;
120 
121 	/*
122 	 * Linked list running through all cgroups that can
123 	 * potentially be reaped by the release agent. Protected by
124 	 * release_list_lock
125 	 */
126 	struct list_head release_list;
127 };
128 
129 /* A css_set is a structure holding pointers to a set of
130  * cgroup_subsys_state objects. This saves space in the task struct
131  * object and speeds up fork()/exit(), since a single inc/dec and a
132  * list_add()/del() can bump the reference count on the entire
133  * cgroup set for a task.
134  */
135 
136 struct css_set {
137 
138 	/* Reference count */
139 	struct kref ref;
140 
141 	/*
142 	 * List running through all cgroup groups. Protected by
143 	 * css_set_lock
144 	 */
145 	struct list_head list;
146 
147 	/*
148 	 * List running through all tasks using this cgroup
149 	 * group. Protected by css_set_lock
150 	 */
151 	struct list_head tasks;
152 
153 	/*
154 	 * List of cg_cgroup_link objects on link chains from
155 	 * cgroups referenced from this css_set. Protected by
156 	 * css_set_lock
157 	 */
158 	struct list_head cg_links;
159 
160 	/*
161 	 * Set of subsystem states, one for each subsystem. This array
162 	 * is immutable after creation apart from the init_css_set
163 	 * during subsystem registration (at boot time).
164 	 */
165 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
166 
167 };
168 
169 /* struct cftype:
170  *
171  * The files in the cgroup filesystem mostly have a very simple read/write
172  * handling, some common function will take care of it. Nevertheless some cases
173  * (read tasks) are special and therefore I define this structure for every
174  * kind of file.
175  *
176  *
177  * When reading/writing to a file:
178  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
179  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
180  */
181 
182 #define MAX_CFTYPE_NAME 64
183 struct cftype {
184 	/* By convention, the name should begin with the name of the
185 	 * subsystem, followed by a period */
186 	char name[MAX_CFTYPE_NAME];
187 	int private;
188 	int (*open) (struct inode *inode, struct file *file);
189 	ssize_t (*read) (struct cgroup *cgrp, struct cftype *cft,
190 			 struct file *file,
191 			 char __user *buf, size_t nbytes, loff_t *ppos);
192 	/*
193 	 * read_uint() is a shortcut for the common case of returning a
194 	 * single integer. Use it in place of read()
195 	 */
196 	u64 (*read_uint) (struct cgroup *cgrp, struct cftype *cft);
197 	ssize_t (*write) (struct cgroup *cgrp, struct cftype *cft,
198 			  struct file *file,
199 			  const char __user *buf, size_t nbytes, loff_t *ppos);
200 
201 	/*
202 	 * write_uint() is a shortcut for the common case of accepting
203 	 * a single integer (as parsed by simple_strtoull) from
204 	 * userspace. Use in place of write(); return 0 or error.
205 	 */
206 	int (*write_uint) (struct cgroup *cgrp, struct cftype *cft, u64 val);
207 
208 	int (*release) (struct inode *inode, struct file *file);
209 };
210 
211 struct cgroup_scanner {
212 	struct cgroup *cg;
213 	int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
214 	void (*process_task)(struct task_struct *p,
215 			struct cgroup_scanner *scan);
216 	struct ptr_heap *heap;
217 };
218 
219 /* Add a new file to the given cgroup directory. Should only be
220  * called by subsystems from within a populate() method */
221 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
222 		       const struct cftype *cft);
223 
224 /* Add a set of new files to the given cgroup directory. Should
225  * only be called by subsystems from within a populate() method */
226 int cgroup_add_files(struct cgroup *cgrp,
227 			struct cgroup_subsys *subsys,
228 			const struct cftype cft[],
229 			int count);
230 
231 int cgroup_is_removed(const struct cgroup *cgrp);
232 
233 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
234 
235 int cgroup_task_count(const struct cgroup *cgrp);
236 
237 /* Return true if the cgroup is a descendant of the current cgroup */
238 int cgroup_is_descendant(const struct cgroup *cgrp);
239 
240 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
241 
242 struct cgroup_subsys {
243 	struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
244 						  struct cgroup *cgrp);
245 	void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
246 	void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
247 	int (*can_attach)(struct cgroup_subsys *ss,
248 			  struct cgroup *cgrp, struct task_struct *tsk);
249 	void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
250 			struct cgroup *old_cgrp, struct task_struct *tsk);
251 	void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
252 	void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
253 	int (*populate)(struct cgroup_subsys *ss,
254 			struct cgroup *cgrp);
255 	void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
256 	void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
257 	int subsys_id;
258 	int active;
259 	int disabled;
260 	int early_init;
261 #define MAX_CGROUP_TYPE_NAMELEN 32
262 	const char *name;
263 
264 	/* Protected by RCU */
265 	struct cgroupfs_root *root;
266 
267 	struct list_head sibling;
268 
269 	void *private;
270 };
271 
272 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
273 #include <linux/cgroup_subsys.h>
274 #undef SUBSYS
275 
276 static inline struct cgroup_subsys_state *cgroup_subsys_state(
277 	struct cgroup *cgrp, int subsys_id)
278 {
279 	return cgrp->subsys[subsys_id];
280 }
281 
282 static inline struct cgroup_subsys_state *task_subsys_state(
283 	struct task_struct *task, int subsys_id)
284 {
285 	return rcu_dereference(task->cgroups->subsys[subsys_id]);
286 }
287 
288 static inline struct cgroup* task_cgroup(struct task_struct *task,
289 					       int subsys_id)
290 {
291 	return task_subsys_state(task, subsys_id)->cgroup;
292 }
293 
294 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss);
295 
296 /* A cgroup_iter should be treated as an opaque object */
297 struct cgroup_iter {
298 	struct list_head *cg_link;
299 	struct list_head *task;
300 };
301 
302 /* To iterate across the tasks in a cgroup:
303  *
304  * 1) call cgroup_iter_start to intialize an iterator
305  *
306  * 2) call cgroup_iter_next() to retrieve member tasks until it
307  *    returns NULL or until you want to end the iteration
308  *
309  * 3) call cgroup_iter_end() to destroy the iterator.
310  *
311  * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
312  *    - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
313  *      callback, but not while calling the process_task() callback.
314  */
315 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
316 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
317 					struct cgroup_iter *it);
318 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
319 int cgroup_scan_tasks(struct cgroup_scanner *scan);
320 int cgroup_attach_task(struct cgroup *, struct task_struct *);
321 
322 #else /* !CONFIG_CGROUPS */
323 
324 static inline int cgroup_init_early(void) { return 0; }
325 static inline int cgroup_init(void) { return 0; }
326 static inline void cgroup_init_smp(void) {}
327 static inline void cgroup_fork(struct task_struct *p) {}
328 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
329 static inline void cgroup_post_fork(struct task_struct *p) {}
330 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
331 
332 static inline void cgroup_lock(void) {}
333 static inline void cgroup_unlock(void) {}
334 static inline int cgroupstats_build(struct cgroupstats *stats,
335 					struct dentry *dentry)
336 {
337 	return -EINVAL;
338 }
339 
340 #endif /* !CONFIG_CGROUPS */
341 
342 #endif /* _LINUX_CGROUP_H */
343