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>
25 struct cgroup_taskset;
28 struct kernfs_open_file;
30 #define MAX_CGROUP_TYPE_NAMELEN 32
31 #define MAX_CGROUP_ROOT_NAMELEN 64
32 #define MAX_CFTYPE_NAME 64
34 /* define the enumeration of all cgroup subsystems */
35 #define SUBSYS(_x) _x ## _cgrp_id,
36 enum cgroup_subsys_id {
37 #include <linux/cgroup_subsys.h>
42 /* bits in struct cgroup_subsys_state flags field */
44 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
45 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
46 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
49 /* bits in struct cgroup flags field */
51 /* Control Group requires release notifications to userspace */
52 CGRP_NOTIFY_ON_RELEASE,
54 * Clone the parent's configuration when creating a new child
55 * cpuset cgroup. For historical reasons, this option can be
56 * specified at mount time and thus is implemented here.
58 CGRP_CPUSET_CLONE_CHILDREN,
61 /* cgroup_root->flags */
63 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
64 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
65 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
70 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
71 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
72 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
74 /* internal flags, do not use outside cgroup core proper */
75 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
76 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
80 * Per-subsystem/per-cgroup state maintained by the system. This is the
81 * fundamental structural building block that controllers deal with.
83 * Fields marked with "PI:" are public and immutable and may be accessed
84 * directly without synchronization.
86 struct cgroup_subsys_state {
87 /* PI: the cgroup that this css is attached to */
88 struct cgroup *cgroup;
90 /* PI: the cgroup subsystem that this css is attached to */
91 struct cgroup_subsys *ss;
93 /* reference count - access via css_[try]get() and css_put() */
94 struct percpu_ref refcnt;
96 /* PI: the parent css */
97 struct cgroup_subsys_state *parent;
99 /* siblings list anchored at the parent's ->children */
100 struct list_head sibling;
101 struct list_head children;
104 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
105 * matching css can be looked up using css_from_id().
112 * Monotonically increasing unique serial number which defines a
113 * uniform order among all csses. It's guaranteed that all
114 * ->children lists are in the ascending order of ->serial_nr and
115 * used to allow interrupting and resuming iterations.
119 /* percpu_ref killing and RCU release */
120 struct rcu_head rcu_head;
121 struct work_struct destroy_work;
125 * A css_set is a structure holding pointers to a set of
126 * cgroup_subsys_state objects. This saves space in the task struct
127 * object and speeds up fork()/exit(), since a single inc/dec and a
128 * list_add()/del() can bump the reference count on the entire cgroup
132 /* Reference count */
136 * List running through all cgroup groups in the same hash
137 * slot. Protected by css_set_lock
139 struct hlist_node hlist;
142 * Lists running through all tasks using this cgroup group.
143 * mg_tasks lists tasks which belong to this cset but are in the
144 * process of being migrated out or in. Protected by
145 * css_set_rwsem, but, during migration, once tasks are moved to
146 * mg_tasks, it can be read safely while holding cgroup_mutex.
148 struct list_head tasks;
149 struct list_head mg_tasks;
152 * List of cgrp_cset_links pointing at cgroups referenced from this
153 * css_set. Protected by css_set_lock.
155 struct list_head cgrp_links;
157 /* the default cgroup associated with this css_set */
158 struct cgroup *dfl_cgrp;
161 * Set of subsystem states, one for each subsystem. This array is
162 * immutable after creation apart from the init_css_set during
163 * subsystem registration (at boot time).
165 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
168 * List of csets participating in the on-going migration either as
169 * source or destination. Protected by cgroup_mutex.
171 struct list_head mg_preload_node;
172 struct list_head mg_node;
175 * If this cset is acting as the source of migration the following
176 * two fields are set. mg_src_cgrp is the source cgroup of the
177 * on-going migration and mg_dst_cset is the destination cset the
178 * target tasks on this cset should be migrated to. Protected by
181 struct cgroup *mg_src_cgrp;
182 struct css_set *mg_dst_cset;
185 * On the default hierarhcy, ->subsys[ssid] may point to a css
186 * attached to an ancestor instead of the cgroup this css_set is
187 * associated with. The following node is anchored at
188 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
189 * iterate through all css's attached to a given cgroup.
191 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
193 /* For RCU-protected deletion */
194 struct rcu_head rcu_head;
198 /* self css with NULL ->ss, points back to this cgroup */
199 struct cgroup_subsys_state self;
201 unsigned long flags; /* "unsigned long" so bitops work */
204 * idr allocated in-hierarchy ID.
206 * ID 0 is not used, the ID of the root cgroup is always 1, and a
207 * new cgroup will be assigned with a smallest available ID.
209 * Allocating/Removing ID must be protected by cgroup_mutex.
214 * If this cgroup contains any tasks, it contributes one to
215 * populated_cnt. All children with non-zero popuplated_cnt of
216 * their own contribute one. The count is zero iff there's no task
217 * in this cgroup or its subtree.
221 struct kernfs_node *kn; /* cgroup kernfs entry */
222 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
225 * The bitmask of subsystems enabled on the child cgroups.
226 * ->subtree_control is the one configured through
227 * "cgroup.subtree_control" while ->child_subsys_mask is the
228 * effective one which may have more subsystems enabled.
229 * Controller knobs are made available iff it's enabled in
232 unsigned int subtree_control;
233 unsigned int child_subsys_mask;
235 /* Private pointers for each registered subsystem */
236 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
238 struct cgroup_root *root;
241 * List of cgrp_cset_links pointing at css_sets with tasks in this
242 * cgroup. Protected by css_set_lock.
244 struct list_head cset_links;
247 * On the default hierarchy, a css_set for a cgroup with some
248 * susbsys disabled will point to css's which are associated with
249 * the closest ancestor which has the subsys enabled. The
250 * following lists all css_sets which point to this cgroup's css
251 * for the given subsystem.
253 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
256 * list of pidlists, up to two for each namespace (one for procs, one
257 * for tasks); created on demand.
259 struct list_head pidlists;
260 struct mutex pidlist_mutex;
262 /* used to wait for offlining of csses */
263 wait_queue_head_t offline_waitq;
265 /* used to schedule release agent */
266 struct work_struct release_agent_work;
270 * A cgroup_root represents the root of a cgroup hierarchy, and may be
271 * associated with a kernfs_root to form an active hierarchy. This is
272 * internal to cgroup core. Don't access directly from controllers.
275 struct kernfs_root *kf_root;
277 /* The bitmask of subsystems attached to this hierarchy */
278 unsigned int subsys_mask;
280 /* Unique id for this hierarchy. */
283 /* The root cgroup. Root is destroyed on its release. */
286 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
289 /* A list running through the active hierarchies */
290 struct list_head root_list;
292 /* Hierarchy-specific flags */
295 /* IDs for cgroups in this hierarchy */
296 struct idr cgroup_idr;
298 /* The path to use for release notifications. */
299 char release_agent_path[PATH_MAX];
301 /* The name for this hierarchy - may be empty */
302 char name[MAX_CGROUP_ROOT_NAMELEN];
306 * struct cftype: handler definitions for cgroup control files
308 * When reading/writing to a file:
309 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
310 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
314 * By convention, the name should begin with the name of the
315 * subsystem, followed by a period. Zero length string indicates
316 * end of cftype array.
318 char name[MAX_CFTYPE_NAME];
321 * If not 0, file mode is set to this value, otherwise it will
322 * be figured out automatically
327 * The maximum length of string, excluding trailing nul, that can
328 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
330 size_t max_write_len;
336 * Fields used for internal bookkeeping. Initialized automatically
337 * during registration.
339 struct cgroup_subsys *ss; /* NULL for cgroup core files */
340 struct list_head node; /* anchored at ss->cfts */
341 struct kernfs_ops *kf_ops;
344 * read_u64() is a shortcut for the common case of returning a
345 * single integer. Use it in place of read()
347 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
349 * read_s64() is a signed version of read_u64()
351 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
353 /* generic seq_file read interface */
354 int (*seq_show)(struct seq_file *sf, void *v);
356 /* optional ops, implement all or none */
357 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
358 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
359 void (*seq_stop)(struct seq_file *sf, void *v);
362 * write_u64() is a shortcut for the common case of accepting
363 * a single integer (as parsed by simple_strtoull) from
364 * userspace. Use in place of write(); return 0 or error.
366 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
369 * write_s64() is a signed version of write_u64()
371 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
375 * write() is the generic write callback which maps directly to
376 * kernfs write operation and overrides all other operations.
377 * Maximum write size is determined by ->max_write_len. Use
378 * of_css/cft() to access the associated css and cft.
380 ssize_t (*write)(struct kernfs_open_file *of,
381 char *buf, size_t nbytes, loff_t off);
383 #ifdef CONFIG_DEBUG_LOCK_ALLOC
384 struct lock_class_key lockdep_key;
389 * Control Group subsystem type.
390 * See Documentation/cgroups/cgroups.txt for details
392 struct cgroup_subsys {
393 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
394 int (*css_online)(struct cgroup_subsys_state *css);
395 void (*css_offline)(struct cgroup_subsys_state *css);
396 void (*css_released)(struct cgroup_subsys_state *css);
397 void (*css_free)(struct cgroup_subsys_state *css);
398 void (*css_reset)(struct cgroup_subsys_state *css);
399 void (*css_e_css_changed)(struct cgroup_subsys_state *css);
401 int (*can_attach)(struct cgroup_subsys_state *css,
402 struct cgroup_taskset *tset);
403 void (*cancel_attach)(struct cgroup_subsys_state *css,
404 struct cgroup_taskset *tset);
405 void (*attach)(struct cgroup_subsys_state *css,
406 struct cgroup_taskset *tset);
407 void (*fork)(struct task_struct *task);
408 void (*exit)(struct cgroup_subsys_state *css,
409 struct cgroup_subsys_state *old_css,
410 struct task_struct *task);
411 void (*bind)(struct cgroup_subsys_state *root_css);
417 * If %false, this subsystem is properly hierarchical -
418 * configuration, resource accounting and restriction on a parent
419 * cgroup cover those of its children. If %true, hierarchy support
420 * is broken in some ways - some subsystems ignore hierarchy
421 * completely while others are only implemented half-way.
423 * It's now disallowed to create nested cgroups if the subsystem is
424 * broken and cgroup core will emit a warning message on such
425 * cases. Eventually, all subsystems will be made properly
426 * hierarchical and this will go away.
428 bool broken_hierarchy;
429 bool warned_broken_hierarchy;
431 /* the following two fields are initialized automtically during boot */
435 /* link to parent, protected by cgroup_lock() */
436 struct cgroup_root *root;
438 /* idr for css->id */
442 * List of cftypes. Each entry is the first entry of an array
443 * terminated by zero length name.
445 struct list_head cfts;
448 * Base cftypes which are automatically registered. The two can
449 * point to the same array.
451 struct cftype *dfl_cftypes; /* for the default hierarchy */
452 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
455 * A subsystem may depend on other subsystems. When such subsystem
456 * is enabled on a cgroup, the depended-upon subsystems are enabled
457 * together if available. Subsystems enabled due to dependency are
458 * not visible to userland until explicitly enabled. The following
459 * specifies the mask of subsystems that this one depends on.
461 unsigned int depends_on;
464 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
467 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
470 * Called from threadgroup_change_begin() and allows cgroup operations to
471 * synchronize against threadgroup changes using a percpu_rw_semaphore.
473 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
475 percpu_down_read(&cgroup_threadgroup_rwsem);
479 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
482 * Called from threadgroup_change_end(). Counterpart of
483 * cgroup_threadcgroup_change_begin().
485 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
487 percpu_up_read(&cgroup_threadgroup_rwsem);
490 #else /* CONFIG_CGROUPS */
492 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
493 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
495 #endif /* CONFIG_CGROUPS */
497 #endif /* _LINUX_CGROUP_DEFS_H */