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
;
31 #define MAX_CGROUP_TYPE_NAMELEN 32
32 #define MAX_CGROUP_ROOT_NAMELEN 64
33 #define MAX_CFTYPE_NAME 64
35 /* define the enumeration of all cgroup subsystems */
36 #define SUBSYS(_x) _x ## _cgrp_id,
37 enum cgroup_subsys_id
{
38 #include <linux/cgroup_subsys.h>
43 /* bits in struct cgroup_subsys_state flags field */
45 CSS_NO_REF
= (1 << 0), /* no reference counting for this css */
46 CSS_ONLINE
= (1 << 1), /* between ->css_online() and ->css_offline() */
47 CSS_RELEASED
= (1 << 2), /* refcnt reached zero, released */
48 CSS_VISIBLE
= (1 << 3), /* css is visible to userland */
51 /* bits in struct cgroup flags field */
53 /* Control Group requires release notifications to userspace */
54 CGRP_NOTIFY_ON_RELEASE
,
56 * Clone the parent's configuration when creating a new child
57 * cpuset cgroup. For historical reasons, this option can be
58 * specified at mount time and thus is implemented here.
60 CGRP_CPUSET_CLONE_CHILDREN
,
63 /* cgroup_root->flags */
65 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
66 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
71 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
72 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
73 CFTYPE_NO_PREFIX
= (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
74 CFTYPE_WORLD_WRITABLE
= (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
76 /* internal flags, do not use outside cgroup core proper */
77 __CFTYPE_ONLY_ON_DFL
= (1 << 16), /* only on default hierarchy */
78 __CFTYPE_NOT_ON_DFL
= (1 << 17), /* not on default hierarchy */
82 * cgroup_file is the handle for a file instance created in a cgroup which
83 * is used, for example, to generate file changed notifications. This can
84 * be obtained by setting cftype->file_offset.
87 /* do not access any fields from outside cgroup core */
88 struct kernfs_node
*kn
;
92 * Per-subsystem/per-cgroup state maintained by the system. This is the
93 * fundamental structural building block that controllers deal with.
95 * Fields marked with "PI:" are public and immutable and may be accessed
96 * directly without synchronization.
98 struct cgroup_subsys_state
{
99 /* PI: the cgroup that this css is attached to */
100 struct cgroup
*cgroup
;
102 /* PI: the cgroup subsystem that this css is attached to */
103 struct cgroup_subsys
*ss
;
105 /* reference count - access via css_[try]get() and css_put() */
106 struct percpu_ref refcnt
;
108 /* PI: the parent css */
109 struct cgroup_subsys_state
*parent
;
111 /* siblings list anchored at the parent's ->children */
112 struct list_head sibling
;
113 struct list_head children
;
116 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
117 * matching css can be looked up using css_from_id().
124 * Monotonically increasing unique serial number which defines a
125 * uniform order among all csses. It's guaranteed that all
126 * ->children lists are in the ascending order of ->serial_nr and
127 * used to allow interrupting and resuming iterations.
132 * Incremented by online self and children. Used to guarantee that
133 * parents are not offlined before their children.
137 /* percpu_ref killing and RCU release */
138 struct rcu_head rcu_head
;
139 struct work_struct destroy_work
;
143 * A css_set is a structure holding pointers to a set of
144 * cgroup_subsys_state objects. This saves space in the task struct
145 * object and speeds up fork()/exit(), since a single inc/dec and a
146 * list_add()/del() can bump the reference count on the entire cgroup
150 /* Reference count */
154 * List running through all cgroup groups in the same hash
155 * slot. Protected by css_set_lock
157 struct hlist_node hlist
;
160 * Lists running through all tasks using this cgroup group.
161 * mg_tasks lists tasks which belong to this cset but are in the
162 * process of being migrated out or in. Protected by
163 * css_set_rwsem, but, during migration, once tasks are moved to
164 * mg_tasks, it can be read safely while holding cgroup_mutex.
166 struct list_head tasks
;
167 struct list_head mg_tasks
;
170 * List of cgrp_cset_links pointing at cgroups referenced from this
171 * css_set. Protected by css_set_lock.
173 struct list_head cgrp_links
;
175 /* the default cgroup associated with this css_set */
176 struct cgroup
*dfl_cgrp
;
179 * Set of subsystem states, one for each subsystem. This array is
180 * immutable after creation apart from the init_css_set during
181 * subsystem registration (at boot time).
183 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
186 * List of csets participating in the on-going migration either as
187 * source or destination. Protected by cgroup_mutex.
189 struct list_head mg_preload_node
;
190 struct list_head mg_node
;
193 * If this cset is acting as the source of migration the following
194 * two fields are set. mg_src_cgrp is the source cgroup of the
195 * on-going migration and mg_dst_cset is the destination cset the
196 * target tasks on this cset should be migrated to. Protected by
199 struct cgroup
*mg_src_cgrp
;
200 struct css_set
*mg_dst_cset
;
203 * On the default hierarhcy, ->subsys[ssid] may point to a css
204 * attached to an ancestor instead of the cgroup this css_set is
205 * associated with. The following node is anchored at
206 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
207 * iterate through all css's attached to a given cgroup.
209 struct list_head e_cset_node
[CGROUP_SUBSYS_COUNT
];
211 /* all css_task_iters currently walking this cset */
212 struct list_head task_iters
;
214 /* For RCU-protected deletion */
215 struct rcu_head rcu_head
;
219 /* self css with NULL ->ss, points back to this cgroup */
220 struct cgroup_subsys_state self
;
222 unsigned long flags
; /* "unsigned long" so bitops work */
225 * idr allocated in-hierarchy ID.
227 * ID 0 is not used, the ID of the root cgroup is always 1, and a
228 * new cgroup will be assigned with a smallest available ID.
230 * Allocating/Removing ID must be protected by cgroup_mutex.
235 * The depth this cgroup is at. The root is at depth zero and each
236 * step down the hierarchy increments the level. This along with
237 * ancestor_ids[] can determine whether a given cgroup is a
238 * descendant of another without traversing the hierarchy.
243 * Each non-empty css_set associated with this cgroup contributes
244 * one to populated_cnt. All children with non-zero popuplated_cnt
245 * of their own contribute one. The count is zero iff there's no
246 * task in this cgroup or its subtree.
250 struct kernfs_node
*kn
; /* cgroup kernfs entry */
251 struct cgroup_file procs_file
; /* handle for "cgroup.procs" */
252 struct cgroup_file events_file
; /* handle for "cgroup.events" */
255 * The bitmask of subsystems enabled on the child cgroups.
256 * ->subtree_control is the one configured through
257 * "cgroup.subtree_control" while ->child_ss_mask is the effective
258 * one which may have more subsystems enabled. Controller knobs
259 * are made available iff it's enabled in ->subtree_control.
263 u16 old_subtree_control
;
264 u16 old_subtree_ss_mask
;
266 /* Private pointers for each registered subsystem */
267 struct cgroup_subsys_state __rcu
*subsys
[CGROUP_SUBSYS_COUNT
];
269 struct cgroup_root
*root
;
272 * List of cgrp_cset_links pointing at css_sets with tasks in this
273 * cgroup. Protected by css_set_lock.
275 struct list_head cset_links
;
278 * On the default hierarchy, a css_set for a cgroup with some
279 * susbsys disabled will point to css's which are associated with
280 * the closest ancestor which has the subsys enabled. The
281 * following lists all css_sets which point to this cgroup's css
282 * for the given subsystem.
284 struct list_head e_csets
[CGROUP_SUBSYS_COUNT
];
287 * list of pidlists, up to two for each namespace (one for procs, one
288 * for tasks); created on demand.
290 struct list_head pidlists
;
291 struct mutex pidlist_mutex
;
293 /* used to wait for offlining of csses */
294 wait_queue_head_t offline_waitq
;
296 /* used to schedule release agent */
297 struct work_struct release_agent_work
;
299 /* ids of the ancestors at each level including self */
304 * A cgroup_root represents the root of a cgroup hierarchy, and may be
305 * associated with a kernfs_root to form an active hierarchy. This is
306 * internal to cgroup core. Don't access directly from controllers.
309 struct kernfs_root
*kf_root
;
311 /* The bitmask of subsystems attached to this hierarchy */
312 unsigned int subsys_mask
;
314 /* Unique id for this hierarchy. */
317 /* The root cgroup. Root is destroyed on its release. */
320 /* for cgrp->ancestor_ids[0] */
321 int cgrp_ancestor_id_storage
;
323 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
326 /* A list running through the active hierarchies */
327 struct list_head root_list
;
329 /* Hierarchy-specific flags */
332 /* IDs for cgroups in this hierarchy */
333 struct idr cgroup_idr
;
335 /* The path to use for release notifications. */
336 char release_agent_path
[PATH_MAX
];
338 /* The name for this hierarchy - may be empty */
339 char name
[MAX_CGROUP_ROOT_NAMELEN
];
343 * struct cftype: handler definitions for cgroup control files
345 * When reading/writing to a file:
346 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
347 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
351 * By convention, the name should begin with the name of the
352 * subsystem, followed by a period. Zero length string indicates
353 * end of cftype array.
355 char name
[MAX_CFTYPE_NAME
];
356 unsigned long private;
359 * The maximum length of string, excluding trailing nul, that can
360 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
362 size_t max_write_len
;
368 * If non-zero, should contain the offset from the start of css to
369 * a struct cgroup_file field. cgroup will record the handle of
370 * the created file into it. The recorded handle can be used as
371 * long as the containing css remains accessible.
373 unsigned int file_offset
;
376 * Fields used for internal bookkeeping. Initialized automatically
377 * during registration.
379 struct cgroup_subsys
*ss
; /* NULL for cgroup core files */
380 struct list_head node
; /* anchored at ss->cfts */
381 struct kernfs_ops
*kf_ops
;
384 * read_u64() is a shortcut for the common case of returning a
385 * single integer. Use it in place of read()
387 u64 (*read_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
389 * read_s64() is a signed version of read_u64()
391 s64 (*read_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
393 /* generic seq_file read interface */
394 int (*seq_show
)(struct seq_file
*sf
, void *v
);
396 /* optional ops, implement all or none */
397 void *(*seq_start
)(struct seq_file
*sf
, loff_t
*ppos
);
398 void *(*seq_next
)(struct seq_file
*sf
, void *v
, loff_t
*ppos
);
399 void (*seq_stop
)(struct seq_file
*sf
, void *v
);
402 * write_u64() is a shortcut for the common case of accepting
403 * a single integer (as parsed by simple_strtoull) from
404 * userspace. Use in place of write(); return 0 or error.
406 int (*write_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
409 * write_s64() is a signed version of write_u64()
411 int (*write_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
415 * write() is the generic write callback which maps directly to
416 * kernfs write operation and overrides all other operations.
417 * Maximum write size is determined by ->max_write_len. Use
418 * of_css/cft() to access the associated css and cft.
420 ssize_t (*write
)(struct kernfs_open_file
*of
,
421 char *buf
, size_t nbytes
, loff_t off
);
423 #ifdef CONFIG_DEBUG_LOCK_ALLOC
424 struct lock_class_key lockdep_key
;
429 * Control Group subsystem type.
430 * See Documentation/cgroups/cgroups.txt for details
432 struct cgroup_subsys
{
433 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup_subsys_state
*parent_css
);
434 int (*css_online
)(struct cgroup_subsys_state
*css
);
435 void (*css_offline
)(struct cgroup_subsys_state
*css
);
436 void (*css_released
)(struct cgroup_subsys_state
*css
);
437 void (*css_free
)(struct cgroup_subsys_state
*css
);
438 void (*css_reset
)(struct cgroup_subsys_state
*css
);
440 int (*can_attach
)(struct cgroup_taskset
*tset
);
441 void (*cancel_attach
)(struct cgroup_taskset
*tset
);
442 void (*attach
)(struct cgroup_taskset
*tset
);
443 int (*can_fork
)(struct task_struct
*task
);
444 void (*cancel_fork
)(struct task_struct
*task
);
445 void (*fork
)(struct task_struct
*task
);
446 void (*exit
)(struct task_struct
*task
);
447 void (*free
)(struct task_struct
*task
);
448 void (*bind
)(struct cgroup_subsys_state
*root_css
);
453 * If %false, this subsystem is properly hierarchical -
454 * configuration, resource accounting and restriction on a parent
455 * cgroup cover those of its children. If %true, hierarchy support
456 * is broken in some ways - some subsystems ignore hierarchy
457 * completely while others are only implemented half-way.
459 * It's now disallowed to create nested cgroups if the subsystem is
460 * broken and cgroup core will emit a warning message on such
461 * cases. Eventually, all subsystems will be made properly
462 * hierarchical and this will go away.
464 bool broken_hierarchy
:1;
465 bool warned_broken_hierarchy
:1;
467 /* the following two fields are initialized automtically during boot */
471 /* optional, initialized automatically during boot if not set */
472 const char *legacy_name
;
474 /* link to parent, protected by cgroup_lock() */
475 struct cgroup_root
*root
;
477 /* idr for css->id */
481 * List of cftypes. Each entry is the first entry of an array
482 * terminated by zero length name.
484 struct list_head cfts
;
487 * Base cftypes which are automatically registered. The two can
488 * point to the same array.
490 struct cftype
*dfl_cftypes
; /* for the default hierarchy */
491 struct cftype
*legacy_cftypes
; /* for the legacy hierarchies */
494 * A subsystem may depend on other subsystems. When such subsystem
495 * is enabled on a cgroup, the depended-upon subsystems are enabled
496 * together if available. Subsystems enabled due to dependency are
497 * not visible to userland until explicitly enabled. The following
498 * specifies the mask of subsystems that this one depends on.
500 unsigned int depends_on
;
503 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
506 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
509 * Called from threadgroup_change_begin() and allows cgroup operations to
510 * synchronize against threadgroup changes using a percpu_rw_semaphore.
512 static inline void cgroup_threadgroup_change_begin(struct task_struct
*tsk
)
514 percpu_down_read(&cgroup_threadgroup_rwsem
);
518 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
521 * Called from threadgroup_change_end(). Counterpart of
522 * cgroup_threadcgroup_change_begin().
524 static inline void cgroup_threadgroup_change_end(struct task_struct
*tsk
)
526 percpu_up_read(&cgroup_threadgroup_rwsem
);
529 #else /* CONFIG_CGROUPS */
531 #define CGROUP_SUBSYS_COUNT 0
533 static inline void cgroup_threadgroup_change_begin(struct task_struct
*tsk
) {}
534 static inline void cgroup_threadgroup_change_end(struct task_struct
*tsk
) {}
536 #endif /* CONFIG_CGROUPS */
538 #ifdef CONFIG_SOCK_CGROUP_DATA
541 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
542 * per-socket cgroup information except for memcg association.
544 * On legacy hierarchies, net_prio and net_cls controllers directly set
545 * attributes on each sock which can then be tested by the network layer.
546 * On the default hierarchy, each sock is associated with the cgroup it was
547 * created in and the networking layer can match the cgroup directly.
549 * To avoid carrying all three cgroup related fields separately in sock,
550 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
551 * On boot, sock_cgroup_data records the cgroup that the sock was created
552 * in so that cgroup2 matches can be made; however, once either net_prio or
553 * net_cls starts being used, the area is overriden to carry prioidx and/or
554 * classid. The two modes are distinguished by whether the lowest bit is
555 * set. Clear bit indicates cgroup pointer while set bit prioidx and
558 * While userland may start using net_prio or net_cls at any time, once
559 * either is used, cgroup2 matching no longer works. There is no reason to
560 * mix the two and this is in line with how legacy and v2 compatibility is
561 * handled. On mode switch, cgroup references which are already being
562 * pointed to by socks may be leaked. While this can be remedied by adding
563 * synchronization around sock_cgroup_data, given that the number of leaked
564 * cgroups is bound and highly unlikely to be high, this seems to be the
567 struct sock_cgroup_data
{
569 #ifdef __LITTLE_ENDIAN
589 * There's a theoretical window where the following accessors race with
590 * updaters and return part of the previous pointer as the prioidx or
591 * classid. Such races are short-lived and the result isn't critical.
593 static inline u16
sock_cgroup_prioidx(struct sock_cgroup_data
*skcd
)
595 /* fallback to 1 which is always the ID of the root cgroup */
596 return (skcd
->is_data
& 1) ? skcd
->prioidx
: 1;
599 static inline u32
sock_cgroup_classid(struct sock_cgroup_data
*skcd
)
601 /* fallback to 0 which is the unconfigured default classid */
602 return (skcd
->is_data
& 1) ? skcd
->classid
: 0;
606 * If invoked concurrently, the updaters may clobber each other. The
607 * caller is responsible for synchronization.
609 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data
*skcd
,
612 struct sock_cgroup_data skcd_buf
= {{ .val
= READ_ONCE(skcd
->val
) }};
614 if (sock_cgroup_prioidx(&skcd_buf
) == prioidx
)
617 if (!(skcd_buf
.is_data
& 1)) {
619 skcd_buf
.is_data
= 1;
622 skcd_buf
.prioidx
= prioidx
;
623 WRITE_ONCE(skcd
->val
, skcd_buf
.val
); /* see sock_cgroup_ptr() */
626 static inline void sock_cgroup_set_classid(struct sock_cgroup_data
*skcd
,
629 struct sock_cgroup_data skcd_buf
= {{ .val
= READ_ONCE(skcd
->val
) }};
631 if (sock_cgroup_classid(&skcd_buf
) == classid
)
634 if (!(skcd_buf
.is_data
& 1)) {
636 skcd_buf
.is_data
= 1;
639 skcd_buf
.classid
= classid
;
640 WRITE_ONCE(skcd
->val
, skcd_buf
.val
); /* see sock_cgroup_ptr() */
643 #else /* CONFIG_SOCK_CGROUP_DATA */
645 struct sock_cgroup_data
{
648 #endif /* CONFIG_SOCK_CGROUP_DATA */
650 #endif /* _LINUX_CGROUP_DEFS_H */