2 * linux/cgroup-defs.h - basic definitions for cgroup
4 * This file provides basic type and interface. Include this file directly
5 * only if necessary to avoid cyclic dependencies.
7 #ifndef _LINUX_CGROUP_DEFS_H
8 #define _LINUX_CGROUP_DEFS_H
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 #include <linux/swork.h>
26 struct cgroup_taskset;
29 struct kernfs_open_file;
32 #define MAX_CGROUP_TYPE_NAMELEN 32
33 #define MAX_CGROUP_ROOT_NAMELEN 64
34 #define MAX_CFTYPE_NAME 64
36 /* define the enumeration of all cgroup subsystems */
37 #define SUBSYS(_x) _x ## _cgrp_id,
38 #define SUBSYS_TAG(_t) CGROUP_ ## _t, \
39 __unused_tag_ ## _t = CGROUP_ ## _t - 1,
40 enum cgroup_subsys_id {
41 #include <linux/cgroup_subsys.h>
47 #define CGROUP_CANFORK_COUNT (CGROUP_CANFORK_END - CGROUP_CANFORK_START)
49 /* bits in struct cgroup_subsys_state flags field */
51 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
52 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
53 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
56 /* bits in struct cgroup flags field */
58 /* Control Group requires release notifications to userspace */
59 CGRP_NOTIFY_ON_RELEASE,
61 * Clone the parent's configuration when creating a new child
62 * cpuset cgroup. For historical reasons, this option can be
63 * specified at mount time and thus is implemented here.
65 CGRP_CPUSET_CLONE_CHILDREN,
68 /* cgroup_root->flags */
70 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
71 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
72 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
77 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
78 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
79 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
80 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
82 /* internal flags, do not use outside cgroup core proper */
83 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
84 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
88 * cgroup_file is the handle for a file instance created in a cgroup which
89 * is used, for example, to generate file changed notifications. This can
90 * be obtained by setting cftype->file_offset.
93 /* do not access any fields from outside cgroup core */
94 struct kernfs_node *kn;
98 * Per-subsystem/per-cgroup state maintained by the system. This is the
99 * fundamental structural building block that controllers deal with.
101 * Fields marked with "PI:" are public and immutable and may be accessed
102 * directly without synchronization.
104 struct cgroup_subsys_state {
105 /* PI: the cgroup that this css is attached to */
106 struct cgroup *cgroup;
108 /* PI: the cgroup subsystem that this css is attached to */
109 struct cgroup_subsys *ss;
111 /* reference count - access via css_[try]get() and css_put() */
112 struct percpu_ref refcnt;
114 /* PI: the parent css */
115 struct cgroup_subsys_state *parent;
117 /* siblings list anchored at the parent's ->children */
118 struct list_head sibling;
119 struct list_head children;
122 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
123 * matching css can be looked up using css_from_id().
130 * Monotonically increasing unique serial number which defines a
131 * uniform order among all csses. It's guaranteed that all
132 * ->children lists are in the ascending order of ->serial_nr and
133 * used to allow interrupting and resuming iterations.
138 * Incremented by online self and children. Used to guarantee that
139 * parents are not offlined before their children.
143 /* percpu_ref killing and RCU release */
144 struct rcu_head rcu_head;
145 struct work_struct destroy_work;
146 struct swork_event destroy_swork;
150 * A css_set is a structure holding pointers to a set of
151 * cgroup_subsys_state objects. This saves space in the task struct
152 * object and speeds up fork()/exit(), since a single inc/dec and a
153 * list_add()/del() can bump the reference count on the entire cgroup
157 /* Reference count */
161 * List running through all cgroup groups in the same hash
162 * slot. Protected by css_set_lock
164 struct hlist_node hlist;
167 * Lists running through all tasks using this cgroup group.
168 * mg_tasks lists tasks which belong to this cset but are in the
169 * process of being migrated out or in. Protected by
170 * css_set_rwsem, but, during migration, once tasks are moved to
171 * mg_tasks, it can be read safely while holding cgroup_mutex.
173 struct list_head tasks;
174 struct list_head mg_tasks;
177 * List of cgrp_cset_links pointing at cgroups referenced from this
178 * css_set. Protected by css_set_lock.
180 struct list_head cgrp_links;
182 /* the default cgroup associated with this css_set */
183 struct cgroup *dfl_cgrp;
186 * Set of subsystem states, one for each subsystem. This array is
187 * immutable after creation apart from the init_css_set during
188 * subsystem registration (at boot time).
190 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
193 * List of csets participating in the on-going migration either as
194 * source or destination. Protected by cgroup_mutex.
196 struct list_head mg_preload_node;
197 struct list_head mg_node;
200 * If this cset is acting as the source of migration the following
201 * two fields are set. mg_src_cgrp is the source cgroup of the
202 * on-going migration and mg_dst_cset is the destination cset the
203 * target tasks on this cset should be migrated to. Protected by
206 struct cgroup *mg_src_cgrp;
207 struct css_set *mg_dst_cset;
210 * On the default hierarhcy, ->subsys[ssid] may point to a css
211 * attached to an ancestor instead of the cgroup this css_set is
212 * associated with. The following node is anchored at
213 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
214 * iterate through all css's attached to a given cgroup.
216 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
218 /* all css_task_iters currently walking this cset */
219 struct list_head task_iters;
221 /* dead and being drained, ignore for migration */
224 /* For RCU-protected deletion */
225 struct rcu_head rcu_head;
229 /* self css with NULL ->ss, points back to this cgroup */
230 struct cgroup_subsys_state self;
232 unsigned long flags; /* "unsigned long" so bitops work */
235 * idr allocated in-hierarchy ID.
237 * ID 0 is not used, the ID of the root cgroup is always 1, and a
238 * new cgroup will be assigned with a smallest available ID.
240 * Allocating/Removing ID must be protected by cgroup_mutex.
245 * Each non-empty css_set associated with this cgroup contributes
246 * one to populated_cnt. All children with non-zero popuplated_cnt
247 * of their own contribute one. The count is zero iff there's no
248 * task in this cgroup or its subtree.
252 struct kernfs_node *kn; /* cgroup kernfs entry */
253 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
254 struct cgroup_file events_file; /* handle for "cgroup.events" */
257 * The bitmask of subsystems enabled on the child cgroups.
258 * ->subtree_control is the one configured through
259 * "cgroup.subtree_control" while ->child_subsys_mask is the
260 * effective one which may have more subsystems enabled.
261 * Controller knobs are made available iff it's enabled in
264 unsigned int subtree_control;
265 unsigned int child_subsys_mask;
267 /* Private pointers for each registered subsystem */
268 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
270 struct cgroup_root *root;
273 * List of cgrp_cset_links pointing at css_sets with tasks in this
274 * cgroup. Protected by css_set_lock.
276 struct list_head cset_links;
279 * On the default hierarchy, a css_set for a cgroup with some
280 * susbsys disabled will point to css's which are associated with
281 * the closest ancestor which has the subsys enabled. The
282 * following lists all css_sets which point to this cgroup's css
283 * for the given subsystem.
285 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
288 * list of pidlists, up to two for each namespace (one for procs, one
289 * for tasks); created on demand.
291 struct list_head pidlists;
292 struct mutex pidlist_mutex;
294 /* used to wait for offlining of csses */
295 wait_queue_head_t offline_waitq;
297 /* used to schedule release agent */
298 struct work_struct release_agent_work;
302 * A cgroup_root represents the root of a cgroup hierarchy, and may be
303 * associated with a kernfs_root to form an active hierarchy. This is
304 * internal to cgroup core. Don't access directly from controllers.
307 struct kernfs_root *kf_root;
309 /* The bitmask of subsystems attached to this hierarchy */
310 unsigned int subsys_mask;
312 /* Unique id for this hierarchy. */
315 /* The root cgroup. Root is destroyed on its release. */
318 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
321 /* A list running through the active hierarchies */
322 struct list_head root_list;
324 /* Hierarchy-specific flags */
327 /* IDs for cgroups in this hierarchy */
328 struct idr cgroup_idr;
330 /* The path to use for release notifications. */
331 char release_agent_path[PATH_MAX];
333 /* The name for this hierarchy - may be empty */
334 char name[MAX_CGROUP_ROOT_NAMELEN];
338 * struct cftype: handler definitions for cgroup control files
340 * When reading/writing to a file:
341 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
342 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
346 * By convention, the name should begin with the name of the
347 * subsystem, followed by a period. Zero length string indicates
348 * end of cftype array.
350 char name[MAX_CFTYPE_NAME];
351 unsigned long private;
354 * The maximum length of string, excluding trailing nul, that can
355 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
357 size_t max_write_len;
363 * If non-zero, should contain the offset from the start of css to
364 * a struct cgroup_file field. cgroup will record the handle of
365 * the created file into it. The recorded handle can be used as
366 * long as the containing css remains accessible.
368 unsigned int file_offset;
371 * Fields used for internal bookkeeping. Initialized automatically
372 * during registration.
374 struct cgroup_subsys *ss; /* NULL for cgroup core files */
375 struct list_head node; /* anchored at ss->cfts */
376 struct kernfs_ops *kf_ops;
379 * read_u64() is a shortcut for the common case of returning a
380 * single integer. Use it in place of read()
382 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
384 * read_s64() is a signed version of read_u64()
386 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
388 /* generic seq_file read interface */
389 int (*seq_show)(struct seq_file *sf, void *v);
391 /* optional ops, implement all or none */
392 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
393 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
394 void (*seq_stop)(struct seq_file *sf, void *v);
397 * write_u64() is a shortcut for the common case of accepting
398 * a single integer (as parsed by simple_strtoull) from
399 * userspace. Use in place of write(); return 0 or error.
401 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
404 * write_s64() is a signed version of write_u64()
406 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
410 * write() is the generic write callback which maps directly to
411 * kernfs write operation and overrides all other operations.
412 * Maximum write size is determined by ->max_write_len. Use
413 * of_css/cft() to access the associated css and cft.
415 ssize_t (*write)(struct kernfs_open_file *of,
416 char *buf, size_t nbytes, loff_t off);
418 #ifdef CONFIG_DEBUG_LOCK_ALLOC
419 struct lock_class_key lockdep_key;
424 * Control Group subsystem type.
425 * See Documentation/cgroups/cgroups.txt for details
427 struct cgroup_subsys {
428 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
429 int (*css_online)(struct cgroup_subsys_state *css);
430 void (*css_offline)(struct cgroup_subsys_state *css);
431 void (*css_released)(struct cgroup_subsys_state *css);
432 void (*css_free)(struct cgroup_subsys_state *css);
433 void (*css_reset)(struct cgroup_subsys_state *css);
434 void (*css_e_css_changed)(struct cgroup_subsys_state *css);
436 int (*can_attach)(struct cgroup_taskset *tset);
437 void (*cancel_attach)(struct cgroup_taskset *tset);
438 void (*attach)(struct cgroup_taskset *tset);
439 void (*post_attach)(void);
440 int (*can_fork)(struct task_struct *task, void **priv_p);
441 void (*cancel_fork)(struct task_struct *task, void *priv);
442 void (*fork)(struct task_struct *task, void *priv);
443 void (*exit)(struct task_struct *task);
444 void (*free)(struct task_struct *task);
445 void (*bind)(struct cgroup_subsys_state *root_css);
450 * If %false, this subsystem is properly hierarchical -
451 * configuration, resource accounting and restriction on a parent
452 * cgroup cover those of its children. If %true, hierarchy support
453 * is broken in some ways - some subsystems ignore hierarchy
454 * completely while others are only implemented half-way.
456 * It's now disallowed to create nested cgroups if the subsystem is
457 * broken and cgroup core will emit a warning message on such
458 * cases. Eventually, all subsystems will be made properly
459 * hierarchical and this will go away.
461 bool broken_hierarchy;
462 bool warned_broken_hierarchy;
464 /* the following two fields are initialized automtically during boot */
468 /* optional, initialized automatically during boot if not set */
469 const char *legacy_name;
471 /* link to parent, protected by cgroup_lock() */
472 struct cgroup_root *root;
474 /* idr for css->id */
478 * List of cftypes. Each entry is the first entry of an array
479 * terminated by zero length name.
481 struct list_head cfts;
484 * Base cftypes which are automatically registered. The two can
485 * point to the same array.
487 struct cftype *dfl_cftypes; /* for the default hierarchy */
488 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
491 * A subsystem may depend on other subsystems. When such subsystem
492 * is enabled on a cgroup, the depended-upon subsystems are enabled
493 * together if available. Subsystems enabled due to dependency are
494 * not visible to userland until explicitly enabled. The following
495 * specifies the mask of subsystems that this one depends on.
497 unsigned int depends_on;
500 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
503 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
506 * Called from threadgroup_change_begin() and allows cgroup operations to
507 * synchronize against threadgroup changes using a percpu_rw_semaphore.
509 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
511 percpu_down_read(&cgroup_threadgroup_rwsem);
515 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
518 * Called from threadgroup_change_end(). Counterpart of
519 * cgroup_threadcgroup_change_begin().
521 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
523 percpu_up_read(&cgroup_threadgroup_rwsem);
526 #else /* CONFIG_CGROUPS */
528 #define CGROUP_CANFORK_COUNT 0
529 #define CGROUP_SUBSYS_COUNT 0
531 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
532 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
534 #endif /* CONFIG_CGROUPS */
536 #endif /* _LINUX_CGROUP_DEFS_H */