xref: /linux-6.15/include/linux/cgroup.h (revision 930cc144)
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/cpumask.h>
13 #include <linux/nodemask.h>
14 #include <linux/rcupdate.h>
15 #include <linux/cgroupstats.h>
16 #include <linux/prio_heap.h>
17 #include <linux/rwsem.h>
18 
19 #ifdef CONFIG_CGROUPS
20 
21 struct cgroupfs_root;
22 struct cgroup_subsys;
23 struct inode;
24 struct cgroup;
25 
26 extern int cgroup_init_early(void);
27 extern int cgroup_init(void);
28 extern void cgroup_init_smp(void);
29 extern void cgroup_lock(void);
30 extern bool cgroup_lock_live_group(struct cgroup *cgrp);
31 extern void cgroup_unlock(void);
32 extern void cgroup_fork(struct task_struct *p);
33 extern void cgroup_fork_callbacks(struct task_struct *p);
34 extern void cgroup_post_fork(struct task_struct *p);
35 extern void cgroup_exit(struct task_struct *p, int run_callbacks);
36 extern int cgroupstats_build(struct cgroupstats *stats,
37 				struct dentry *dentry);
38 
39 extern struct file_operations proc_cgroup_operations;
40 
41 /* Define the enumeration of all cgroup subsystems */
42 #define SUBSYS(_x) _x ## _subsys_id,
43 enum cgroup_subsys_id {
44 #include <linux/cgroup_subsys.h>
45 	CGROUP_SUBSYS_COUNT
46 };
47 #undef SUBSYS
48 
49 /* Per-subsystem/per-cgroup state maintained by the system. */
50 struct cgroup_subsys_state {
51 	/* The cgroup that this subsystem is attached to. Useful
52 	 * for subsystems that want to know about the cgroup
53 	 * hierarchy structure */
54 	struct cgroup *cgroup;
55 
56 	/* State maintained by the cgroup system to allow
57 	 * subsystems to be "busy". Should be accessed via css_get()
58 	 * and css_put() */
59 
60 	atomic_t refcnt;
61 
62 	unsigned long flags;
63 };
64 
65 /* bits in struct cgroup_subsys_state flags field */
66 enum {
67 	CSS_ROOT, /* This CSS is the root of the subsystem */
68 };
69 
70 /*
71  * Call css_get() to hold a reference on the cgroup;
72  *
73  */
74 
75 static inline void css_get(struct cgroup_subsys_state *css)
76 {
77 	/* We don't need to reference count the root state */
78 	if (!test_bit(CSS_ROOT, &css->flags))
79 		atomic_inc(&css->refcnt);
80 }
81 /*
82  * css_put() should be called to release a reference taken by
83  * css_get()
84  */
85 
86 extern void __css_put(struct cgroup_subsys_state *css);
87 static inline void css_put(struct cgroup_subsys_state *css)
88 {
89 	if (!test_bit(CSS_ROOT, &css->flags))
90 		__css_put(css);
91 }
92 
93 /* bits in struct cgroup flags field */
94 enum {
95 	/* Control Group is dead */
96 	CGRP_REMOVED,
97 	/* Control Group has previously had a child cgroup or a task,
98 	 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
99 	CGRP_RELEASABLE,
100 	/* Control Group requires release notifications to userspace */
101 	CGRP_NOTIFY_ON_RELEASE,
102 };
103 
104 struct cgroup {
105 	unsigned long flags;		/* "unsigned long" so bitops work */
106 
107 	/* count users of this cgroup. >0 means busy, but doesn't
108 	 * necessarily indicate the number of tasks in the
109 	 * cgroup */
110 	atomic_t count;
111 
112 	/*
113 	 * We link our 'sibling' struct into our parent's 'children'.
114 	 * Our children link their 'sibling' into our 'children'.
115 	 */
116 	struct list_head sibling;	/* my parent's children */
117 	struct list_head children;	/* my children */
118 
119 	struct cgroup *parent;	/* my parent */
120 	struct dentry *dentry;	  	/* cgroup fs entry */
121 
122 	/* Private pointers for each registered subsystem */
123 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
124 
125 	struct cgroupfs_root *root;
126 	struct cgroup *top_cgroup;
127 
128 	/*
129 	 * List of cg_cgroup_links pointing at css_sets with
130 	 * tasks in this cgroup. Protected by css_set_lock
131 	 */
132 	struct list_head css_sets;
133 
134 	/*
135 	 * Linked list running through all cgroups that can
136 	 * potentially be reaped by the release agent. Protected by
137 	 * release_list_lock
138 	 */
139 	struct list_head release_list;
140 
141 	/* pids_mutex protects the fields below */
142 	struct rw_semaphore pids_mutex;
143 	/* Array of process ids in the cgroup */
144 	pid_t *tasks_pids;
145 	/* How many files are using the current tasks_pids array */
146 	int pids_use_count;
147 	/* Length of the current tasks_pids array */
148 	int pids_length;
149 };
150 
151 /* A css_set is a structure holding pointers to a set of
152  * cgroup_subsys_state objects. This saves space in the task struct
153  * object and speeds up fork()/exit(), since a single inc/dec and a
154  * list_add()/del() can bump the reference count on the entire
155  * cgroup set for a task.
156  */
157 
158 struct css_set {
159 
160 	/* Reference count */
161 	atomic_t refcount;
162 
163 	/*
164 	 * List running through all cgroup groups in the same hash
165 	 * slot. Protected by css_set_lock
166 	 */
167 	struct hlist_node hlist;
168 
169 	/*
170 	 * List running through all tasks using this cgroup
171 	 * group. Protected by css_set_lock
172 	 */
173 	struct list_head tasks;
174 
175 	/*
176 	 * List of cg_cgroup_link objects on link chains from
177 	 * cgroups referenced from this css_set. Protected by
178 	 * css_set_lock
179 	 */
180 	struct list_head cg_links;
181 
182 	/*
183 	 * Set of subsystem states, one for each subsystem. This array
184 	 * is immutable after creation apart from the init_css_set
185 	 * during subsystem registration (at boot time).
186 	 */
187 	struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
188 };
189 
190 /*
191  * cgroup_map_cb is an abstract callback API for reporting map-valued
192  * control files
193  */
194 
195 struct cgroup_map_cb {
196 	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
197 	void *state;
198 };
199 
200 /* struct cftype:
201  *
202  * The files in the cgroup filesystem mostly have a very simple read/write
203  * handling, some common function will take care of it. Nevertheless some cases
204  * (read tasks) are special and therefore I define this structure for every
205  * kind of file.
206  *
207  *
208  * When reading/writing to a file:
209  *	- the cgroup to use is file->f_dentry->d_parent->d_fsdata
210  *	- the 'cftype' of the file is file->f_dentry->d_fsdata
211  */
212 
213 #define MAX_CFTYPE_NAME 64
214 struct cftype {
215 	/* By convention, the name should begin with the name of the
216 	 * subsystem, followed by a period */
217 	char name[MAX_CFTYPE_NAME];
218 	int private;
219 
220 	/*
221 	 * If non-zero, defines the maximum length of string that can
222 	 * be passed to write_string; defaults to 64
223 	 */
224 	size_t max_write_len;
225 
226 	int (*open)(struct inode *inode, struct file *file);
227 	ssize_t (*read)(struct cgroup *cgrp, struct cftype *cft,
228 			struct file *file,
229 			char __user *buf, size_t nbytes, loff_t *ppos);
230 	/*
231 	 * read_u64() is a shortcut for the common case of returning a
232 	 * single integer. Use it in place of read()
233 	 */
234 	u64 (*read_u64)(struct cgroup *cgrp, struct cftype *cft);
235 	/*
236 	 * read_s64() is a signed version of read_u64()
237 	 */
238 	s64 (*read_s64)(struct cgroup *cgrp, struct cftype *cft);
239 	/*
240 	 * read_map() is used for defining a map of key/value
241 	 * pairs. It should call cb->fill(cb, key, value) for each
242 	 * entry. The key/value pairs (and their ordering) should not
243 	 * change between reboots.
244 	 */
245 	int (*read_map)(struct cgroup *cont, struct cftype *cft,
246 			struct cgroup_map_cb *cb);
247 	/*
248 	 * read_seq_string() is used for outputting a simple sequence
249 	 * using seqfile.
250 	 */
251 	int (*read_seq_string)(struct cgroup *cont, struct cftype *cft,
252 			       struct seq_file *m);
253 
254 	ssize_t (*write)(struct cgroup *cgrp, struct cftype *cft,
255 			 struct file *file,
256 			 const char __user *buf, size_t nbytes, loff_t *ppos);
257 
258 	/*
259 	 * write_u64() is a shortcut for the common case of accepting
260 	 * a single integer (as parsed by simple_strtoull) from
261 	 * userspace. Use in place of write(); return 0 or error.
262 	 */
263 	int (*write_u64)(struct cgroup *cgrp, struct cftype *cft, u64 val);
264 	/*
265 	 * write_s64() is a signed version of write_u64()
266 	 */
267 	int (*write_s64)(struct cgroup *cgrp, struct cftype *cft, s64 val);
268 
269 	/*
270 	 * write_string() is passed a nul-terminated kernelspace
271 	 * buffer of maximum length determined by max_write_len.
272 	 * Returns 0 or -ve error code.
273 	 */
274 	int (*write_string)(struct cgroup *cgrp, struct cftype *cft,
275 			    const char *buffer);
276 	/*
277 	 * trigger() callback can be used to get some kick from the
278 	 * userspace, when the actual string written is not important
279 	 * at all. The private field can be used to determine the
280 	 * kick type for multiplexing.
281 	 */
282 	int (*trigger)(struct cgroup *cgrp, unsigned int event);
283 
284 	int (*release)(struct inode *inode, struct file *file);
285 };
286 
287 struct cgroup_scanner {
288 	struct cgroup *cg;
289 	int (*test_task)(struct task_struct *p, struct cgroup_scanner *scan);
290 	void (*process_task)(struct task_struct *p,
291 			struct cgroup_scanner *scan);
292 	struct ptr_heap *heap;
293 };
294 
295 /* Add a new file to the given cgroup directory. Should only be
296  * called by subsystems from within a populate() method */
297 int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
298 		       const struct cftype *cft);
299 
300 /* Add a set of new files to the given cgroup directory. Should
301  * only be called by subsystems from within a populate() method */
302 int cgroup_add_files(struct cgroup *cgrp,
303 			struct cgroup_subsys *subsys,
304 			const struct cftype cft[],
305 			int count);
306 
307 int cgroup_is_removed(const struct cgroup *cgrp);
308 
309 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen);
310 
311 int cgroup_task_count(const struct cgroup *cgrp);
312 
313 /* Return true if the cgroup is a descendant of the current cgroup */
314 int cgroup_is_descendant(const struct cgroup *cgrp);
315 
316 /* Control Group subsystem type. See Documentation/cgroups.txt for details */
317 
318 struct cgroup_subsys {
319 	struct cgroup_subsys_state *(*create)(struct cgroup_subsys *ss,
320 						  struct cgroup *cgrp);
321 	void (*pre_destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
322 	void (*destroy)(struct cgroup_subsys *ss, struct cgroup *cgrp);
323 	int (*can_attach)(struct cgroup_subsys *ss,
324 			  struct cgroup *cgrp, struct task_struct *tsk);
325 	void (*attach)(struct cgroup_subsys *ss, struct cgroup *cgrp,
326 			struct cgroup *old_cgrp, struct task_struct *tsk);
327 	void (*fork)(struct cgroup_subsys *ss, struct task_struct *task);
328 	void (*exit)(struct cgroup_subsys *ss, struct task_struct *task);
329 	int (*populate)(struct cgroup_subsys *ss,
330 			struct cgroup *cgrp);
331 	void (*post_clone)(struct cgroup_subsys *ss, struct cgroup *cgrp);
332 	void (*bind)(struct cgroup_subsys *ss, struct cgroup *root);
333 	/*
334 	 * This routine is called with the task_lock of mm->owner held
335 	 */
336 	void (*mm_owner_changed)(struct cgroup_subsys *ss,
337 					struct cgroup *old,
338 					struct cgroup *new,
339 					struct task_struct *p);
340 	int subsys_id;
341 	int active;
342 	int disabled;
343 	int early_init;
344 #define MAX_CGROUP_TYPE_NAMELEN 32
345 	const char *name;
346 
347 	/* Protected by RCU */
348 	struct cgroupfs_root *root;
349 
350 	struct list_head sibling;
351 
352 	void *private;
353 };
354 
355 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _subsys;
356 #include <linux/cgroup_subsys.h>
357 #undef SUBSYS
358 
359 static inline struct cgroup_subsys_state *cgroup_subsys_state(
360 	struct cgroup *cgrp, int subsys_id)
361 {
362 	return cgrp->subsys[subsys_id];
363 }
364 
365 static inline struct cgroup_subsys_state *task_subsys_state(
366 	struct task_struct *task, int subsys_id)
367 {
368 	return rcu_dereference(task->cgroups->subsys[subsys_id]);
369 }
370 
371 static inline struct cgroup* task_cgroup(struct task_struct *task,
372 					       int subsys_id)
373 {
374 	return task_subsys_state(task, subsys_id)->cgroup;
375 }
376 
377 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *ss,
378 							char *nodename);
379 
380 /* A cgroup_iter should be treated as an opaque object */
381 struct cgroup_iter {
382 	struct list_head *cg_link;
383 	struct list_head *task;
384 };
385 
386 /* To iterate across the tasks in a cgroup:
387  *
388  * 1) call cgroup_iter_start to intialize an iterator
389  *
390  * 2) call cgroup_iter_next() to retrieve member tasks until it
391  *    returns NULL or until you want to end the iteration
392  *
393  * 3) call cgroup_iter_end() to destroy the iterator.
394  *
395  * Or, call cgroup_scan_tasks() to iterate through every task in a cpuset.
396  *    - cgroup_scan_tasks() holds the css_set_lock when calling the test_task()
397  *      callback, but not while calling the process_task() callback.
398  */
399 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it);
400 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
401 					struct cgroup_iter *it);
402 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it);
403 int cgroup_scan_tasks(struct cgroup_scanner *scan);
404 int cgroup_attach_task(struct cgroup *, struct task_struct *);
405 
406 void cgroup_mm_owner_callbacks(struct task_struct *old,
407 			       struct task_struct *new);
408 
409 #else /* !CONFIG_CGROUPS */
410 
411 static inline int cgroup_init_early(void) { return 0; }
412 static inline int cgroup_init(void) { return 0; }
413 static inline void cgroup_init_smp(void) {}
414 static inline void cgroup_fork(struct task_struct *p) {}
415 static inline void cgroup_fork_callbacks(struct task_struct *p) {}
416 static inline void cgroup_post_fork(struct task_struct *p) {}
417 static inline void cgroup_exit(struct task_struct *p, int callbacks) {}
418 
419 static inline void cgroup_lock(void) {}
420 static inline void cgroup_unlock(void) {}
421 static inline int cgroupstats_build(struct cgroupstats *stats,
422 					struct dentry *dentry)
423 {
424 	return -EINVAL;
425 }
426 
427 static inline void cgroup_mm_owner_callbacks(struct task_struct *old,
428 					     struct task_struct *new) {}
429 
430 #endif /* !CONFIG_CGROUPS */
431 
432 #endif /* _LINUX_CGROUP_H */
433