1 /*
2 * linux/cgroup-defs.h - basic definitions for cgroup
3 *
4 * This file provides basic type and interface. Include this file directly
5 * only if necessary to avoid cyclic dependencies.
6 */
7 #ifndef _LINUX_CGROUP_DEFS_H
8 #define _LINUX_CGROUP_DEFS_H
9
10 #include <linux/limits.h>
11 #include <linux/list.h>
12 #include <linux/idr.h>
13 #include <linux/wait.h>
14 #include <linux/mutex.h>
15 #include <linux/rcupdate.h>
16 #include <linux/percpu-refcount.h>
17 #include <linux/percpu-rwsem.h>
18 #include <linux/workqueue.h>
19
20 #ifdef CONFIG_CGROUPS
21
22 struct cgroup;
23 struct cgroup_root;
24 struct cgroup_subsys;
25 struct cgroup_taskset;
26 struct kernfs_node;
27 struct kernfs_ops;
28 struct kernfs_open_file;
29 struct seq_file;
30
31 #define MAX_CGROUP_TYPE_NAMELEN 32
32 #define MAX_CGROUP_ROOT_NAMELEN 64
33 #define MAX_CFTYPE_NAME 64
34
35 /* define the enumeration of all cgroup subsystems */
36 #define SUBSYS(_x) _x ## _cgrp_id,
37 #define SUBSYS_TAG(_t) CGROUP_ ## _t, \
38 __unused_tag_ ## _t = CGROUP_ ## _t - 1,
39 enum cgroup_subsys_id {
40 #include <linux/cgroup_subsys.h>
41 CGROUP_SUBSYS_COUNT,
42 };
43 #undef SUBSYS_TAG
44 #undef SUBSYS
45
46 #define CGROUP_CANFORK_COUNT (CGROUP_CANFORK_END - CGROUP_CANFORK_START)
47
48 /* bits in struct cgroup_subsys_state flags field */
49 enum {
50 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
51 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
52 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
53 };
54
55 /* bits in struct cgroup flags field */
56 enum {
57 /* Control Group requires release notifications to userspace */
58 CGRP_NOTIFY_ON_RELEASE,
59 /*
60 * Clone the parent's configuration when creating a new child
61 * cpuset cgroup. For historical reasons, this option can be
62 * specified at mount time and thus is implemented here.
63 */
64 CGRP_CPUSET_CLONE_CHILDREN,
65 };
66
67 /* cgroup_root->flags */
68 enum {
69 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
70 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
71 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
72 };
73
74 /* cftype->flags */
75 enum {
76 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
77 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
78 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
79 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
80
81 /* internal flags, do not use outside cgroup core proper */
82 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
83 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
84 };
85
86 /*
87 * cgroup_file is the handle for a file instance created in a cgroup which
88 * is used, for example, to generate file changed notifications. This can
89 * be obtained by setting cftype->file_offset.
90 */
91 struct cgroup_file {
92 /* do not access any fields from outside cgroup core */
93 struct kernfs_node *kn;
94 };
95
96 /*
97 * Per-subsystem/per-cgroup state maintained by the system. This is the
98 * fundamental structural building block that controllers deal with.
99 *
100 * Fields marked with "PI:" are public and immutable and may be accessed
101 * directly without synchronization.
102 */
103 struct cgroup_subsys_state {
104 /* PI: the cgroup that this css is attached to */
105 struct cgroup *cgroup;
106
107 /* PI: the cgroup subsystem that this css is attached to */
108 struct cgroup_subsys *ss;
109
110 /* reference count - access via css_[try]get() and css_put() */
111 struct percpu_ref refcnt;
112
113 /* PI: the parent css */
114 struct cgroup_subsys_state *parent;
115
116 /* siblings list anchored at the parent's ->children */
117 struct list_head sibling;
118 struct list_head children;
119
120 /*
121 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
122 * matching css can be looked up using css_from_id().
123 */
124 int id;
125
126 unsigned int flags;
127
128 /*
129 * Monotonically increasing unique serial number which defines a
130 * uniform order among all csses. It's guaranteed that all
131 * ->children lists are in the ascending order of ->serial_nr and
132 * used to allow interrupting and resuming iterations.
133 */
134 u64 serial_nr;
135
136 /* percpu_ref killing and RCU release */
137 struct rcu_head rcu_head;
138 struct work_struct destroy_work;
139 };
140
141 /*
142 * A css_set is a structure holding pointers to a set of
143 * cgroup_subsys_state objects. This saves space in the task struct
144 * object and speeds up fork()/exit(), since a single inc/dec and a
145 * list_add()/del() can bump the reference count on the entire cgroup
146 * set for a task.
147 */
148 struct css_set {
149 /* Reference count */
150 atomic_t refcount;
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 * Lists running through all tasks using this cgroup group.
160 * mg_tasks lists tasks which belong to this cset but are in the
161 * process of being migrated out or in. Protected by
162 * css_set_rwsem, but, during migration, once tasks are moved to
163 * mg_tasks, it can be read safely while holding cgroup_mutex.
164 */
165 struct list_head tasks;
166 struct list_head mg_tasks;
167
168 /*
169 * List of cgrp_cset_links pointing at cgroups referenced from this
170 * css_set. Protected by css_set_lock.
171 */
172 struct list_head cgrp_links;
173
174 /* the default cgroup associated with this css_set */
175 struct cgroup *dfl_cgrp;
176
177 /*
178 * Set of subsystem states, one for each subsystem. This array is
179 * immutable after creation apart from the init_css_set during
180 * subsystem registration (at boot time).
181 */
182 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
183
184 /*
185 * List of csets participating in the on-going migration either as
186 * source or destination. Protected by cgroup_mutex.
187 */
188 struct list_head mg_preload_node;
189 struct list_head mg_node;
190
191 /*
192 * If this cset is acting as the source of migration the following
193 * two fields are set. mg_src_cgrp is the source cgroup of the
194 * on-going migration and mg_dst_cset is the destination cset the
195 * target tasks on this cset should be migrated to. Protected by
196 * cgroup_mutex.
197 */
198 struct cgroup *mg_src_cgrp;
199 struct css_set *mg_dst_cset;
200
201 /*
202 * On the default hierarhcy, ->subsys[ssid] may point to a css
203 * attached to an ancestor instead of the cgroup this css_set is
204 * associated with. The following node is anchored at
205 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
206 * iterate through all css's attached to a given cgroup.
207 */
208 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
209
210 /* all css_task_iters currently walking this cset */
211 struct list_head task_iters;
212
213 /* For RCU-protected deletion */
214 struct rcu_head rcu_head;
215 };
216
217 struct cgroup {
218 /* self css with NULL ->ss, points back to this cgroup */
219 struct cgroup_subsys_state self;
220
221 unsigned long flags; /* "unsigned long" so bitops work */
222
223 /*
224 * idr allocated in-hierarchy ID.
225 *
226 * ID 0 is not used, the ID of the root cgroup is always 1, and a
227 * new cgroup will be assigned with a smallest available ID.
228 *
229 * Allocating/Removing ID must be protected by cgroup_mutex.
230 */
231 int id;
232
233 /*
234 * Each non-empty css_set associated with this cgroup contributes
235 * one to populated_cnt. All children with non-zero popuplated_cnt
236 * of their own contribute one. The count is zero iff there's no
237 * task in this cgroup or its subtree.
238 */
239 int populated_cnt;
240
241 struct kernfs_node *kn; /* cgroup kernfs entry */
242 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
243 struct cgroup_file events_file; /* handle for "cgroup.events" */
244
245 /*
246 * The bitmask of subsystems enabled on the child cgroups.
247 * ->subtree_control is the one configured through
248 * "cgroup.subtree_control" while ->child_subsys_mask is the
249 * effective one which may have more subsystems enabled.
250 * Controller knobs are made available iff it's enabled in
251 * ->subtree_control.
252 */
253 unsigned int subtree_control;
254 unsigned int child_subsys_mask;
255
256 /* Private pointers for each registered subsystem */
257 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
258
259 struct cgroup_root *root;
260
261 /*
262 * List of cgrp_cset_links pointing at css_sets with tasks in this
263 * cgroup. Protected by css_set_lock.
264 */
265 struct list_head cset_links;
266
267 /*
268 * On the default hierarchy, a css_set for a cgroup with some
269 * susbsys disabled will point to css's which are associated with
270 * the closest ancestor which has the subsys enabled. The
271 * following lists all css_sets which point to this cgroup's css
272 * for the given subsystem.
273 */
274 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
275
276 /*
277 * list of pidlists, up to two for each namespace (one for procs, one
278 * for tasks); created on demand.
279 */
280 struct list_head pidlists;
281 struct mutex pidlist_mutex;
282
283 /* used to wait for offlining of csses */
284 wait_queue_head_t offline_waitq;
285
286 /* used to schedule release agent */
287 struct work_struct release_agent_work;
288 };
289
290 /*
291 * A cgroup_root represents the root of a cgroup hierarchy, and may be
292 * associated with a kernfs_root to form an active hierarchy. This is
293 * internal to cgroup core. Don't access directly from controllers.
294 */
295 struct cgroup_root {
296 struct kernfs_root *kf_root;
297
298 /* The bitmask of subsystems attached to this hierarchy */
299 unsigned int subsys_mask;
300
301 /* Unique id for this hierarchy. */
302 int hierarchy_id;
303
304 /* The root cgroup. Root is destroyed on its release. */
305 struct cgroup cgrp;
306
307 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
308 atomic_t nr_cgrps;
309
310 /* A list running through the active hierarchies */
311 struct list_head root_list;
312
313 /* Hierarchy-specific flags */
314 unsigned int flags;
315
316 /* IDs for cgroups in this hierarchy */
317 struct idr cgroup_idr;
318
319 /* The path to use for release notifications. */
320 char release_agent_path[PATH_MAX];
321
322 /* The name for this hierarchy - may be empty */
323 char name[MAX_CGROUP_ROOT_NAMELEN];
324 };
325
326 /*
327 * struct cftype: handler definitions for cgroup control files
328 *
329 * When reading/writing to a file:
330 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
331 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
332 */
333 struct cftype {
334 /*
335 * By convention, the name should begin with the name of the
336 * subsystem, followed by a period. Zero length string indicates
337 * end of cftype array.
338 */
339 char name[MAX_CFTYPE_NAME];
340 unsigned long private;
341
342 /*
343 * The maximum length of string, excluding trailing nul, that can
344 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
345 */
346 size_t max_write_len;
347
348 /* CFTYPE_* flags */
349 unsigned int flags;
350
351 /*
352 * If non-zero, should contain the offset from the start of css to
353 * a struct cgroup_file field. cgroup will record the handle of
354 * the created file into it. The recorded handle can be used as
355 * long as the containing css remains accessible.
356 */
357 unsigned int file_offset;
358
359 /*
360 * Fields used for internal bookkeeping. Initialized automatically
361 * during registration.
362 */
363 struct cgroup_subsys *ss; /* NULL for cgroup core files */
364 struct list_head node; /* anchored at ss->cfts */
365 struct kernfs_ops *kf_ops;
366
367 /*
368 * read_u64() is a shortcut for the common case of returning a
369 * single integer. Use it in place of read()
370 */
371 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
372 /*
373 * read_s64() is a signed version of read_u64()
374 */
375 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
376
377 /* generic seq_file read interface */
378 int (*seq_show)(struct seq_file *sf, void *v);
379
380 /* optional ops, implement all or none */
381 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
382 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
383 void (*seq_stop)(struct seq_file *sf, void *v);
384
385 /*
386 * write_u64() is a shortcut for the common case of accepting
387 * a single integer (as parsed by simple_strtoull) from
388 * userspace. Use in place of write(); return 0 or error.
389 */
390 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
391 u64 val);
392 /*
393 * write_s64() is a signed version of write_u64()
394 */
395 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
396 s64 val);
397
398 /*
399 * write() is the generic write callback which maps directly to
400 * kernfs write operation and overrides all other operations.
401 * Maximum write size is determined by ->max_write_len. Use
402 * of_css/cft() to access the associated css and cft.
403 */
404 ssize_t (*write)(struct kernfs_open_file *of,
405 char *buf, size_t nbytes, loff_t off);
406
407 #ifdef CONFIG_DEBUG_LOCK_ALLOC
408 struct lock_class_key lockdep_key;
409 #endif
410 };
411
412 /*
413 * Control Group subsystem type.
414 * See Documentation/cgroups/cgroups.txt for details
415 */
416 struct cgroup_subsys {
417 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
418 int (*css_online)(struct cgroup_subsys_state *css);
419 void (*css_offline)(struct cgroup_subsys_state *css);
420 void (*css_released)(struct cgroup_subsys_state *css);
421 void (*css_free)(struct cgroup_subsys_state *css);
422 void (*css_reset)(struct cgroup_subsys_state *css);
423 void (*css_e_css_changed)(struct cgroup_subsys_state *css);
424
425 int (*can_attach)(struct cgroup_taskset *tset);
426 void (*cancel_attach)(struct cgroup_taskset *tset);
427 void (*attach)(struct cgroup_taskset *tset);
428 int (*can_fork)(struct task_struct *task, void **priv_p);
429 void (*cancel_fork)(struct task_struct *task, void *priv);
430 void (*fork)(struct task_struct *task, void *priv);
431 void (*exit)(struct task_struct *task);
432 void (*free)(struct task_struct *task);
433 void (*bind)(struct cgroup_subsys_state *root_css);
434
435 int early_init;
436
437 /*
438 * If %false, this subsystem is properly hierarchical -
439 * configuration, resource accounting and restriction on a parent
440 * cgroup cover those of its children. If %true, hierarchy support
441 * is broken in some ways - some subsystems ignore hierarchy
442 * completely while others are only implemented half-way.
443 *
444 * It's now disallowed to create nested cgroups if the subsystem is
445 * broken and cgroup core will emit a warning message on such
446 * cases. Eventually, all subsystems will be made properly
447 * hierarchical and this will go away.
448 */
449 bool broken_hierarchy;
450 bool warned_broken_hierarchy;
451
452 /* the following two fields are initialized automtically during boot */
453 int id;
454 const char *name;
455
456 /* optional, initialized automatically during boot if not set */
457 const char *legacy_name;
458
459 /* link to parent, protected by cgroup_lock() */
460 struct cgroup_root *root;
461
462 /* idr for css->id */
463 struct idr css_idr;
464
465 /*
466 * List of cftypes. Each entry is the first entry of an array
467 * terminated by zero length name.
468 */
469 struct list_head cfts;
470
471 /*
472 * Base cftypes which are automatically registered. The two can
473 * point to the same array.
474 */
475 struct cftype *dfl_cftypes; /* for the default hierarchy */
476 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
477
478 /*
479 * A subsystem may depend on other subsystems. When such subsystem
480 * is enabled on a cgroup, the depended-upon subsystems are enabled
481 * together if available. Subsystems enabled due to dependency are
482 * not visible to userland until explicitly enabled. The following
483 * specifies the mask of subsystems that this one depends on.
484 */
485 unsigned int depends_on;
486 };
487
488 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
489
490 /**
491 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
492 * @tsk: target task
493 *
494 * Called from threadgroup_change_begin() and allows cgroup operations to
495 * synchronize against threadgroup changes using a percpu_rw_semaphore.
496 */
cgroup_threadgroup_change_begin(struct task_struct * tsk)497 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
498 {
499 percpu_down_read(&cgroup_threadgroup_rwsem);
500 }
501
502 /**
503 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
504 * @tsk: target task
505 *
506 * Called from threadgroup_change_end(). Counterpart of
507 * cgroup_threadcgroup_change_begin().
508 */
cgroup_threadgroup_change_end(struct task_struct * tsk)509 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
510 {
511 percpu_up_read(&cgroup_threadgroup_rwsem);
512 }
513
514 #else /* CONFIG_CGROUPS */
515
516 #define CGROUP_CANFORK_COUNT 0
517 #define CGROUP_SUBSYS_COUNT 0
518
cgroup_threadgroup_change_begin(struct task_struct * tsk)519 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
cgroup_threadgroup_change_end(struct task_struct * tsk)520 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
521
522 #endif /* CONFIG_CGROUPS */
523
524 #endif /* _LINUX_CGROUP_DEFS_H */
525