xref: /linux-6.15/include/linux/cgroup.h (revision 2d6ffcca)
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 /* bits in struct cgroup flags field */
92 enum {
93 	/* Control Group is dead */
94 	CGRP_REMOVED,
95 	/* Control Group has previously had a child cgroup or a task,
96 	 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
97 	CGRP_RELEASABLE,
98 	/* Control Group requires release notifications to userspace */
99 	CGRP_NOTIFY_ON_RELEASE,
100 };
101 
102 struct cgroup {
103 	unsigned long flags;		/* "unsigned long" so bitops work */
104 
105 	/* count users of this cgroup. >0 means busy, but doesn't
106 	 * necessarily indicate the number of tasks in the
107 	 * cgroup */
108 	atomic_t count;
109 
110 	/*
111 	 * We link our 'sibling' struct into our parent's 'children'.
112 	 * Our children link their 'sibling' into our 'children'.
113 	 */
114 	struct list_head sibling;	/* my parent's children */
115 	struct list_head children;	/* my children */
116 
117 	struct cgroup *parent;	/* my parent */
118 	struct dentry *dentry;	  	/* cgroup fs entry */
119 
120 	/* Private pointers for each registered subsystem */
121 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
122 
123 	struct cgroupfs_root *root;
124 	struct cgroup *top_cgroup;
125 
126 	/*
127 	 * List of cg_cgroup_links pointing at css_sets with
128 	 * tasks in this cgroup. Protected by css_set_lock
129 	 */
130 	struct list_head css_sets;
131 
132 	/*
133 	 * Linked list running through all cgroups that can
134 	 * potentially be reaped by the release agent. Protected by
135 	 * release_list_lock
136 	 */
137 	struct list_head release_list;
138 };
139 
140 /* A css_set is a structure holding pointers to a set of
141  * cgroup_subsys_state objects. This saves space in the task struct
142  * object and speeds up fork()/exit(), since a single inc/dec and a
143  * list_add()/del() can bump the reference count on the entire
144  * cgroup set for a task.
145  */
146 
147 struct css_set {
148 
149 	/* Reference count */
150 	struct kref ref;
151 
152 	/*
153 	 * List running through all cgroup groups in the same hash
154 	 * slot. Protected by css_set_lock
155 	 */
156 	struct hlist_node hlist;
157 
158 	/*
159 	 * List running through all tasks using this cgroup
160 	 * group. Protected by css_set_lock
161 	 */
162 	struct list_head tasks;
163 
164 	/*
165 	 * List of cg_cgroup_link objects on link chains from
166 	 * cgroups referenced from this css_set. Protected by
167 	 * css_set_lock
168 	 */
169 	struct list_head cg_links;
170 
171 	/*
172 	 * Set of subsystem states, one for each subsystem. This array
173 	 * is immutable after creation apart from the init_css_set
174 	 * during subsystem registration (at boot time).
175 	 */
176 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
177 };
178 
179 /*
180  * cgroup_map_cb is an abstract callback API for reporting map-valued
181  * control files
182  */
183 
184 struct cgroup_map_cb {
185 	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
186 	void *state;
187 };
188 
189 /* struct cftype:
190  *
191  * The files in the cgroup filesystem mostly have a very simple read/write
192  * handling, some common function will take care of it. Nevertheless some cases
193  * (read tasks) are special and therefore I define this structure for every
194  * kind of file.
195  *
196  *
197  * When reading/writing to a file:
198  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
199  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
200  */
201 
202 #define MAX_CFTYPE_NAME 64
203 struct cftype {
204 	/* By convention, the name should begin with the name of the
205 	 * subsystem, followed by a period */
206 	char name[MAX_CFTYPE_NAME];
207 	int private;
208 	int (*open) (struct inode *inode, struct file *file);
209 	ssize_t (*read) (struct cgroup *cgrp, struct cftype *cft,
210 			 struct file *file,
211 			 char __user *buf, size_t nbytes, loff_t *ppos);
212 	/*
213 	 * read_u64() is a shortcut for the common case of returning a
214 	 * single integer. Use it in place of read()
215 	 */
216 	u64 (*read_u64) (struct cgroup *cgrp, struct cftype *cft);
217 	/*
218 	 * read_s64() is a signed version of read_u64()
219 	 */
220 	s64 (*read_s64) (struct cgroup *cgrp, struct cftype *cft);
221 	/*
222 	 * read_map() is used for defining a map of key/value
223 	 * pairs. It should call cb->fill(cb, key, value) for each
224 	 * entry. The key/value pairs (and their ordering) should not
225 	 * change between reboots.
226 	 */
227 	int (*read_map) (struct cgroup *cont, struct cftype *cft,
228 			 struct cgroup_map_cb *cb);
229 	/*
230 	 * read_seq_string() is used for outputting a simple sequence
231 	 * using seqfile.
232 	 */
233 	int (*read_seq_string) (struct cgroup *cont, struct cftype *cft,
234 			 struct seq_file *m);
235 
236 	ssize_t (*write) (struct cgroup *cgrp, struct cftype *cft,
237 			  struct file *file,
238 			  const char __user *buf, size_t nbytes, loff_t *ppos);
239 
240 	/*
241 	 * write_u64() is a shortcut for the common case of accepting
242 	 * a single integer (as parsed by simple_strtoull) from
243 	 * userspace. Use in place of write(); return 0 or error.
244 	 */
245 	int (*write_u64) (struct cgroup *cgrp, struct cftype *cft, u64 val);
246 	/*
247 	 * write_s64() is a signed version of write_u64()
248 	 */
249 	int (*write_s64) (struct cgroup *cgrp, struct cftype *cft, s64 val);
250 
251 	/*
252 	 * trigger() callback can be used to get some kick from the
253 	 * userspace, when the actual string written is not important
254 	 * at all. The private field can be used to determine the
255 	 * kick type for multiplexing.
256 	 */
257 	int (*trigger)(struct cgroup *cgrp, unsigned int event);
258 
259 	int (*release) (struct inode *inode, struct file *file);
260 };
261 
262 struct cgroup_scanner {
263 	struct cgroup *cg;
264 	int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
265 	void (*process_task)(struct task_struct *p,
266 			struct cgroup_scanner *scan);
267 	struct ptr_heap *heap;
268 };
269 
270 /* Add a new file to the given cgroup directory. Should only be
271  * called by subsystems from within a populate() method */
272 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
273 		       const struct cftype *cft);
274 
275 /* Add a set of new files to the given cgroup directory. Should
276  * only be called by subsystems from within a populate() method */
277 int cgroup_add_files(struct cgroup *cgrp,
278 			struct cgroup_subsys *subsys,
279 			const struct cftype cft[],
280 			int count);
281 
282 int cgroup_is_removed(const struct cgroup *cgrp);
283 
284 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
285 
286 int cgroup_task_count(const struct cgroup *cgrp);
287 
288 /* Return true if the cgroup is a descendant of the current cgroup */
289 int cgroup_is_descendant(const struct cgroup *cgrp);
290 
291 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
292 
293 struct cgroup_subsys {
294 	struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
295 						  struct cgroup *cgrp);
296 	void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
297 	void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
298 	int (*can_attach)(struct cgroup_subsys *ss,
299 			  struct cgroup *cgrp, struct task_struct *tsk);
300 	void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
301 			struct cgroup *old_cgrp, struct task_struct *tsk);
302 	void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
303 	void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
304 	int (*populate)(struct cgroup_subsys *ss,
305 			struct cgroup *cgrp);
306 	void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
307 	void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
308 	/*
309 	 * This routine is called with the task_lock of mm->owner held
310 	 */
311 	void (*mm_owner_changed)(struct cgroup_subsys *ss,
312 					struct cgroup *old,
313 					struct cgroup *new);
314 	int subsys_id;
315 	int active;
316 	int disabled;
317 	int early_init;
318 #define MAX_CGROUP_TYPE_NAMELEN 32
319 	const char *name;
320 
321 	/* Protected by RCU */
322 	struct cgroupfs_root *root;
323 
324 	struct list_head sibling;
325 
326 	void *private;
327 };
328 
329 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
330 #include <linux/cgroup_subsys.h>
331 #undef SUBSYS
332 
333 static inline struct cgroup_subsys_state *cgroup_subsys_state(
334 	struct cgroup *cgrp, int subsys_id)
335 {
336 	return cgrp->subsys[subsys_id];
337 }
338 
339 static inline struct cgroup_subsys_state *task_subsys_state(
340 	struct task_struct *task, int subsys_id)
341 {
342 	return rcu_dereference(task->cgroups->subsys[subsys_id]);
343 }
344 
345 static inline struct cgroup* task_cgroup(struct task_struct *task,
346 					       int subsys_id)
347 {
348 	return task_subsys_state(task, subsys_id)->cgroup;
349 }
350 
351 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss);
352 
353 /* A cgroup_iter should be treated as an opaque object */
354 struct cgroup_iter {
355 	struct list_head *cg_link;
356 	struct list_head *task;
357 };
358 
359 /* To iterate across the tasks in a cgroup:
360  *
361  * 1) call cgroup_iter_start to intialize an iterator
362  *
363  * 2) call cgroup_iter_next() to retrieve member tasks until it
364  *    returns NULL or until you want to end the iteration
365  *
366  * 3) call cgroup_iter_end() to destroy the iterator.
367  *
368  * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
369  *    - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
370  *      callback, but not while calling the process_task() callback.
371  */
372 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
373 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
374 					struct cgroup_iter *it);
375 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
376 int cgroup_scan_tasks(struct cgroup_scanner *scan);
377 int cgroup_attach_task(struct cgroup *, struct task_struct *);
378 
379 #else /* !CONFIG_CGROUPS */
380 
381 static inline int cgroup_init_early(void) { return 0; }
382 static inline int cgroup_init(void) { return 0; }
383 static inline void cgroup_init_smp(void) {}
384 static inline void cgroup_fork(struct task_struct *p) {}
385 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
386 static inline void cgroup_post_fork(struct task_struct *p) {}
387 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
388 
389 static inline void cgroup_lock(void) {}
390 static inline void cgroup_unlock(void) {}
391 static inline int cgroupstats_build(struct cgroupstats *stats,
392 					struct dentry *dentry)
393 {
394 	return -EINVAL;
395 }
396 
397 #endif /* !CONFIG_CGROUPS */
398 
399 #ifdef CONFIG_MM_OWNER
400 extern void
401 cgroup_mm_owner_callbacks(struct task_struct *old, struct task_struct *new);
402 #else /* !CONFIG_MM_OWNER */
403 static inline void
404 cgroup_mm_owner_callbacks(struct task_struct *old, struct task_struct *new)
405 {
406 }
407 #endif /* CONFIG_MM_OWNER */
408 #endif /* _LINUX_CGROUP_H */
409