1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/cgroup-defs.h - basic definitions for cgroup
5 * This file provides basic type and interface. Include this file directly
6 * only if necessary to avoid cyclic dependencies.
8 #ifndef _LINUX_CGROUP_DEFS_H
9 #define _LINUX_CGROUP_DEFS_H
11 #include <linux/limits.h>
12 #include <linux/list.h>
13 #include <linux/idr.h>
14 #include <linux/wait.h>
15 #include <linux/mutex.h>
16 #include <linux/rcupdate.h>
17 #include <linux/refcount.h>
18 #include <linux/percpu-refcount.h>
19 #include <linux/percpu-rwsem.h>
20 #include <linux/workqueue.h>
21 #include <linux/bpf-cgroup.h>
22 #include <linux/psi_types.h>
29 struct cgroup_taskset
;
32 struct kernfs_open_file
;
34 struct poll_table_struct
;
36 #define MAX_CGROUP_TYPE_NAMELEN 32
37 #define MAX_CGROUP_ROOT_NAMELEN 64
38 #define MAX_CFTYPE_NAME 64
40 /* define the enumeration of all cgroup subsystems */
41 #define SUBSYS(_x) _x ## _cgrp_id,
42 enum cgroup_subsys_id
{
43 #include <linux/cgroup_subsys.h>
48 /* bits in struct cgroup_subsys_state flags field */
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 CSS_VISIBLE
= (1 << 3), /* css is visible to userland */
54 CSS_DYING
= (1 << 4), /* css is dying */
57 /* bits in struct cgroup flags field */
59 /* Control Group requires release notifications to userspace */
60 CGRP_NOTIFY_ON_RELEASE
,
62 * Clone the parent's configuration when creating a new child
63 * cpuset cgroup. For historical reasons, this option can be
64 * specified at mount time and thus is implemented here.
66 CGRP_CPUSET_CLONE_CHILDREN
,
69 /* cgroup_root->flags */
71 CGRP_ROOT_NOPREFIX
= (1 << 1), /* mounted subsystems have no named prefix */
72 CGRP_ROOT_XATTR
= (1 << 2), /* supports extended attributes */
75 * Consider namespaces as delegation boundaries. If this flag is
76 * set, controller specific interface files in a namespace root
77 * aren't writeable from inside the namespace.
79 CGRP_ROOT_NS_DELEGATE
= (1 << 3),
82 * Enable cpuset controller in v1 cgroup to use v2 behavior.
84 CGRP_ROOT_CPUSET_V2_MODE
= (1 << 4),
89 CFTYPE_ONLY_ON_ROOT
= (1 << 0), /* only create on root cgrp */
90 CFTYPE_NOT_ON_ROOT
= (1 << 1), /* don't create on root cgrp */
91 CFTYPE_NS_DELEGATABLE
= (1 << 2), /* writeable beyond delegation boundaries */
93 CFTYPE_NO_PREFIX
= (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
94 CFTYPE_WORLD_WRITABLE
= (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
96 /* internal flags, do not use outside cgroup core proper */
97 __CFTYPE_ONLY_ON_DFL
= (1 << 16), /* only on default hierarchy */
98 __CFTYPE_NOT_ON_DFL
= (1 << 17), /* not on default hierarchy */
102 * cgroup_file is the handle for a file instance created in a cgroup which
103 * is used, for example, to generate file changed notifications. This can
104 * be obtained by setting cftype->file_offset.
107 /* do not access any fields from outside cgroup core */
108 struct kernfs_node
*kn
;
112 * Per-subsystem/per-cgroup state maintained by the system. This is the
113 * fundamental structural building block that controllers deal with.
115 * Fields marked with "PI:" are public and immutable and may be accessed
116 * directly without synchronization.
118 struct cgroup_subsys_state
{
119 /* PI: the cgroup that this css is attached to */
120 struct cgroup
*cgroup
;
122 /* PI: the cgroup subsystem that this css is attached to */
123 struct cgroup_subsys
*ss
;
125 /* reference count - access via css_[try]get() and css_put() */
126 struct percpu_ref refcnt
;
128 /* siblings list anchored at the parent's ->children */
129 struct list_head sibling
;
130 struct list_head children
;
133 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
134 * matching css can be looked up using css_from_id().
141 * Monotonically increasing unique serial number which defines a
142 * uniform order among all csses. It's guaranteed that all
143 * ->children lists are in the ascending order of ->serial_nr and
144 * used to allow interrupting and resuming iterations.
149 * Incremented by online self and children. Used to guarantee that
150 * parents are not offlined before their children.
154 /* percpu_ref killing and RCU release */
155 struct rcu_head rcu_head
;
156 struct work_struct destroy_work
;
159 * PI: the parent css. Placed here for cache proximity to following
160 * fields of the containing structure.
162 struct cgroup_subsys_state
*parent
;
166 * A css_set is a structure holding pointers to a set of
167 * cgroup_subsys_state objects. This saves space in the task struct
168 * object and speeds up fork()/exit(), since a single inc/dec and a
169 * list_add()/del() can bump the reference count on the entire cgroup
174 * Set of subsystem states, one for each subsystem. This array is
175 * immutable after creation apart from the init_css_set during
176 * subsystem registration (at boot time).
178 struct cgroup_subsys_state
*subsys
[CGROUP_SUBSYS_COUNT
];
180 /* reference count */
184 * For a domain cgroup, the following points to self. If threaded,
185 * to the matching cset of the nearest domain ancestor. The
186 * dom_cset provides access to the domain cgroup and its csses to
187 * which domain level resource consumptions should be charged.
189 struct css_set
*dom_cset
;
191 /* the default cgroup associated with this css_set */
192 struct cgroup
*dfl_cgrp
;
194 /* internal task count, protected by css_set_lock */
198 * Lists running through all tasks using this cgroup group.
199 * mg_tasks lists tasks which belong to this cset but are in the
200 * process of being migrated out or in. Protected by
201 * css_set_rwsem, but, during migration, once tasks are moved to
202 * mg_tasks, it can be read safely while holding cgroup_mutex.
204 struct list_head tasks
;
205 struct list_head mg_tasks
;
207 /* all css_task_iters currently walking this cset */
208 struct list_head task_iters
;
211 * On the default hierarhcy, ->subsys[ssid] may point to a css
212 * attached to an ancestor instead of the cgroup this css_set is
213 * associated with. The following node is anchored at
214 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
215 * iterate through all css's attached to a given cgroup.
217 struct list_head e_cset_node
[CGROUP_SUBSYS_COUNT
];
219 /* all threaded csets whose ->dom_cset points to this cset */
220 struct list_head threaded_csets
;
221 struct list_head threaded_csets_node
;
224 * List running through all cgroup groups in the same hash
225 * slot. Protected by css_set_lock
227 struct hlist_node hlist
;
230 * List of cgrp_cset_links pointing at cgroups referenced from this
231 * css_set. Protected by css_set_lock.
233 struct list_head cgrp_links
;
236 * List of csets participating in the on-going migration either as
237 * source or destination. Protected by cgroup_mutex.
239 struct list_head mg_preload_node
;
240 struct list_head mg_node
;
243 * If this cset is acting as the source of migration the following
244 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
245 * respectively the source and destination cgroups of the on-going
246 * migration. mg_dst_cset is the destination cset the target tasks
247 * on this cset should be migrated to. Protected by cgroup_mutex.
249 struct cgroup
*mg_src_cgrp
;
250 struct cgroup
*mg_dst_cgrp
;
251 struct css_set
*mg_dst_cset
;
253 /* dead and being drained, ignore for migration */
256 /* For RCU-protected deletion */
257 struct rcu_head rcu_head
;
261 /* self css with NULL ->ss, points back to this cgroup */
262 struct cgroup_subsys_state self
;
264 unsigned long flags
; /* "unsigned long" so bitops work */
267 * idr allocated in-hierarchy ID.
269 * ID 0 is not used, the ID of the root cgroup is always 1, and a
270 * new cgroup will be assigned with a smallest available ID.
272 * Allocating/Removing ID must be protected by cgroup_mutex.
277 * The depth this cgroup is at. The root is at depth zero and each
278 * step down the hierarchy increments the level. This along with
279 * ancestor_ids[] can determine whether a given cgroup is a
280 * descendant of another without traversing the hierarchy.
284 /* Maximum allowed descent tree depth */
288 * Keep track of total numbers of visible and dying descent cgroups.
289 * Dying cgroups are cgroups which were deleted by a user,
290 * but are still existing because someone else is holding a reference.
291 * max_descendants is a maximum allowed number of descent cgroups.
294 int nr_dying_descendants
;
298 * Each non-empty css_set associated with this cgroup contributes
299 * one to nr_populated_csets. The counter is zero iff this cgroup
300 * doesn't have any tasks.
302 * All children which have non-zero nr_populated_csets and/or
303 * nr_populated_children of their own contribute one to either
304 * nr_populated_domain_children or nr_populated_threaded_children
305 * depending on their type. Each counter is zero iff all cgroups
306 * of the type in the subtree proper don't have any tasks.
308 int nr_populated_csets
;
309 int nr_populated_domain_children
;
310 int nr_populated_threaded_children
;
312 int nr_threaded_children
; /* # of live threaded child cgroups */
314 struct kernfs_node
*kn
; /* cgroup kernfs entry */
315 struct cgroup_file procs_file
; /* handle for "cgroup.procs" */
316 struct cgroup_file events_file
; /* handle for "cgroup.events" */
319 * The bitmask of subsystems enabled on the child cgroups.
320 * ->subtree_control is the one configured through
321 * "cgroup.subtree_control" while ->child_ss_mask is the effective
322 * one which may have more subsystems enabled. Controller knobs
323 * are made available iff it's enabled in ->subtree_control.
327 u16 old_subtree_control
;
328 u16 old_subtree_ss_mask
;
330 /* Private pointers for each registered subsystem */
331 struct cgroup_subsys_state __rcu
*subsys
[CGROUP_SUBSYS_COUNT
];
333 struct cgroup_root
*root
;
336 * List of cgrp_cset_links pointing at css_sets with tasks in this
337 * cgroup. Protected by css_set_lock.
339 struct list_head cset_links
;
342 * On the default hierarchy, a css_set for a cgroup with some
343 * susbsys disabled will point to css's which are associated with
344 * the closest ancestor which has the subsys enabled. The
345 * following lists all css_sets which point to this cgroup's css
346 * for the given subsystem.
348 struct list_head e_csets
[CGROUP_SUBSYS_COUNT
];
351 * If !threaded, self. If threaded, it points to the nearest
352 * domain ancestor. Inside a threaded subtree, cgroups are exempt
353 * from process granularity and no-internal-task constraint.
354 * Domain level resource consumptions which aren't tied to a
355 * specific task are charged to the dom_cgrp.
357 struct cgroup
*dom_cgrp
;
358 struct cgroup
*old_dom_cgrp
; /* used while enabling threaded */
361 * list of pidlists, up to two for each namespace (one for procs, one
362 * for tasks); created on demand.
364 struct list_head pidlists
;
365 struct mutex pidlist_mutex
;
367 /* used to wait for offlining of csses */
368 wait_queue_head_t offline_waitq
;
370 /* used to schedule release agent */
371 struct work_struct release_agent_work
;
373 /* used to track pressure stalls */
374 struct psi_group psi
;
376 /* used to store eBPF programs */
377 struct cgroup_bpf bpf
;
379 /* ids of the ancestors at each level including self */
384 * A cgroup_root represents the root of a cgroup hierarchy, and may be
385 * associated with a kernfs_root to form an active hierarchy. This is
386 * internal to cgroup core. Don't access directly from controllers.
389 struct kernfs_root
*kf_root
;
391 /* The bitmask of subsystems attached to this hierarchy */
392 unsigned int subsys_mask
;
394 /* Unique id for this hierarchy. */
397 /* The root cgroup. Root is destroyed on its release. */
400 /* for cgrp->ancestor_ids[0] */
401 int cgrp_ancestor_id_storage
;
403 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
406 /* A list running through the active hierarchies */
407 struct list_head root_list
;
409 /* Hierarchy-specific flags */
412 /* IDs for cgroups in this hierarchy */
413 struct idr cgroup_idr
;
415 /* The path to use for release notifications. */
416 char release_agent_path
[PATH_MAX
];
418 /* The name for this hierarchy - may be empty */
419 char name
[MAX_CGROUP_ROOT_NAMELEN
];
423 * struct cftype: handler definitions for cgroup control files
425 * When reading/writing to a file:
426 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
427 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
431 * By convention, the name should begin with the name of the
432 * subsystem, followed by a period. Zero length string indicates
433 * end of cftype array.
435 char name
[MAX_CFTYPE_NAME
];
436 unsigned long private;
439 * The maximum length of string, excluding trailing nul, that can
440 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
442 size_t max_write_len
;
448 * If non-zero, should contain the offset from the start of css to
449 * a struct cgroup_file field. cgroup will record the handle of
450 * the created file into it. The recorded handle can be used as
451 * long as the containing css remains accessible.
453 unsigned int file_offset
;
456 * Fields used for internal bookkeeping. Initialized automatically
457 * during registration.
459 struct cgroup_subsys
*ss
; /* NULL for cgroup core files */
460 struct list_head node
; /* anchored at ss->cfts */
461 struct kernfs_ops
*kf_ops
;
463 int (*open
)(struct kernfs_open_file
*of
);
464 void (*release
)(struct kernfs_open_file
*of
);
467 * read_u64() is a shortcut for the common case of returning a
468 * single integer. Use it in place of read()
470 u64 (*read_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
472 * read_s64() is a signed version of read_u64()
474 s64 (*read_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
);
476 /* generic seq_file read interface */
477 int (*seq_show
)(struct seq_file
*sf
, void *v
);
479 /* optional ops, implement all or none */
480 void *(*seq_start
)(struct seq_file
*sf
, loff_t
*ppos
);
481 void *(*seq_next
)(struct seq_file
*sf
, void *v
, loff_t
*ppos
);
482 void (*seq_stop
)(struct seq_file
*sf
, void *v
);
485 * write_u64() is a shortcut for the common case of accepting
486 * a single integer (as parsed by simple_strtoull) from
487 * userspace. Use in place of write(); return 0 or error.
489 int (*write_u64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
492 * write_s64() is a signed version of write_u64()
494 int (*write_s64
)(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
498 * write() is the generic write callback which maps directly to
499 * kernfs write operation and overrides all other operations.
500 * Maximum write size is determined by ->max_write_len. Use
501 * of_css/cft() to access the associated css and cft.
503 ssize_t (*write
)(struct kernfs_open_file
*of
,
504 char *buf
, size_t nbytes
, loff_t off
);
506 unsigned int (*poll
)(struct kernfs_open_file
*of
,
507 struct poll_table_struct
*pt
);
509 #ifdef CONFIG_DEBUG_LOCK_ALLOC
510 struct lock_class_key lockdep_key
;
515 * Control Group subsystem type.
516 * See Documentation/cgroups/cgroups.txt for details
518 struct cgroup_subsys
{
519 struct cgroup_subsys_state
*(*css_alloc
)(struct cgroup_subsys_state
*parent_css
);
520 int (*css_online
)(struct cgroup_subsys_state
*css
);
521 void (*css_offline
)(struct cgroup_subsys_state
*css
);
522 void (*css_released
)(struct cgroup_subsys_state
*css
);
523 void (*css_free
)(struct cgroup_subsys_state
*css
);
524 void (*css_reset
)(struct cgroup_subsys_state
*css
);
526 int (*can_attach
)(struct cgroup_taskset
*tset
);
527 void (*cancel_attach
)(struct cgroup_taskset
*tset
);
528 void (*attach
)(struct cgroup_taskset
*tset
);
529 void (*post_attach
)(void);
530 int (*can_fork
)(struct task_struct
*task
);
531 void (*cancel_fork
)(struct task_struct
*task
);
532 void (*fork
)(struct task_struct
*task
);
533 void (*exit
)(struct task_struct
*task
);
534 void (*release
)(struct task_struct
*task
);
535 void (*bind
)(struct cgroup_subsys_state
*root_css
);
540 * If %true, the controller, on the default hierarchy, doesn't show
541 * up in "cgroup.controllers" or "cgroup.subtree_control", is
542 * implicitly enabled on all cgroups on the default hierarchy, and
543 * bypasses the "no internal process" constraint. This is for
544 * utility type controllers which is transparent to userland.
546 * An implicit controller can be stolen from the default hierarchy
547 * anytime and thus must be okay with offline csses from previous
548 * hierarchies coexisting with csses for the current one.
550 bool implicit_on_dfl
:1;
553 * If %true, the controller, supports threaded mode on the default
554 * hierarchy. In a threaded subtree, both process granularity and
555 * no-internal-process constraint are ignored and a threaded
556 * controllers should be able to handle that.
558 * Note that as an implicit controller is automatically enabled on
559 * all cgroups on the default hierarchy, it should also be
560 * threaded. implicit && !threaded is not supported.
565 * If %false, this subsystem is properly hierarchical -
566 * configuration, resource accounting and restriction on a parent
567 * cgroup cover those of its children. If %true, hierarchy support
568 * is broken in some ways - some subsystems ignore hierarchy
569 * completely while others are only implemented half-way.
571 * It's now disallowed to create nested cgroups if the subsystem is
572 * broken and cgroup core will emit a warning message on such
573 * cases. Eventually, all subsystems will be made properly
574 * hierarchical and this will go away.
576 bool broken_hierarchy
:1;
577 bool warned_broken_hierarchy
:1;
579 /* the following two fields are initialized automtically during boot */
583 /* optional, initialized automatically during boot if not set */
584 const char *legacy_name
;
586 /* link to parent, protected by cgroup_lock() */
587 struct cgroup_root
*root
;
589 /* idr for css->id */
593 * List of cftypes. Each entry is the first entry of an array
594 * terminated by zero length name.
596 struct list_head cfts
;
599 * Base cftypes which are automatically registered. The two can
600 * point to the same array.
602 struct cftype
*dfl_cftypes
; /* for the default hierarchy */
603 struct cftype
*legacy_cftypes
; /* for the legacy hierarchies */
606 * A subsystem may depend on other subsystems. When such subsystem
607 * is enabled on a cgroup, the depended-upon subsystems are enabled
608 * together if available. Subsystems enabled due to dependency are
609 * not visible to userland until explicitly enabled. The following
610 * specifies the mask of subsystems that this one depends on.
612 unsigned int depends_on
;
615 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
618 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
621 * Allows cgroup operations to synchronize against threadgroup changes
622 * using a percpu_rw_semaphore.
624 static inline void cgroup_threadgroup_change_begin(struct task_struct
*tsk
)
626 percpu_down_read(&cgroup_threadgroup_rwsem
);
630 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
633 * Counterpart of cgroup_threadcgroup_change_begin().
635 static inline void cgroup_threadgroup_change_end(struct task_struct
*tsk
)
637 percpu_up_read(&cgroup_threadgroup_rwsem
);
640 #else /* CONFIG_CGROUPS */
642 #define CGROUP_SUBSYS_COUNT 0
644 static inline void cgroup_threadgroup_change_begin(struct task_struct
*tsk
)
649 static inline void cgroup_threadgroup_change_end(struct task_struct
*tsk
) {}
651 #endif /* CONFIG_CGROUPS */
653 #ifdef CONFIG_SOCK_CGROUP_DATA
656 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
657 * per-socket cgroup information except for memcg association.
659 * On legacy hierarchies, net_prio and net_cls controllers directly set
660 * attributes on each sock which can then be tested by the network layer.
661 * On the default hierarchy, each sock is associated with the cgroup it was
662 * created in and the networking layer can match the cgroup directly.
664 * To avoid carrying all three cgroup related fields separately in sock,
665 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
666 * On boot, sock_cgroup_data records the cgroup that the sock was created
667 * in so that cgroup2 matches can be made; however, once either net_prio or
668 * net_cls starts being used, the area is overriden to carry prioidx and/or
669 * classid. The two modes are distinguished by whether the lowest bit is
670 * set. Clear bit indicates cgroup pointer while set bit prioidx and
673 * While userland may start using net_prio or net_cls at any time, once
674 * either is used, cgroup2 matching no longer works. There is no reason to
675 * mix the two and this is in line with how legacy and v2 compatibility is
676 * handled. On mode switch, cgroup references which are already being
677 * pointed to by socks may be leaked. While this can be remedied by adding
678 * synchronization around sock_cgroup_data, given that the number of leaked
679 * cgroups is bound and highly unlikely to be high, this seems to be the
682 struct sock_cgroup_data
{
684 #ifdef __LITTLE_ENDIAN
704 * There's a theoretical window where the following accessors race with
705 * updaters and return part of the previous pointer as the prioidx or
706 * classid. Such races are short-lived and the result isn't critical.
708 static inline u16
sock_cgroup_prioidx(const struct sock_cgroup_data
*skcd
)
710 /* fallback to 1 which is always the ID of the root cgroup */
711 return (skcd
->is_data
& 1) ? skcd
->prioidx
: 1;
714 static inline u32
sock_cgroup_classid(const struct sock_cgroup_data
*skcd
)
716 /* fallback to 0 which is the unconfigured default classid */
717 return (skcd
->is_data
& 1) ? skcd
->classid
: 0;
721 * If invoked concurrently, the updaters may clobber each other. The
722 * caller is responsible for synchronization.
724 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data
*skcd
,
727 struct sock_cgroup_data skcd_buf
= {{ .val
= READ_ONCE(skcd
->val
) }};
729 if (sock_cgroup_prioidx(&skcd_buf
) == prioidx
)
732 if (!(skcd_buf
.is_data
& 1)) {
734 skcd_buf
.is_data
= 1;
737 skcd_buf
.prioidx
= prioidx
;
738 WRITE_ONCE(skcd
->val
, skcd_buf
.val
); /* see sock_cgroup_ptr() */
741 static inline void sock_cgroup_set_classid(struct sock_cgroup_data
*skcd
,
744 struct sock_cgroup_data skcd_buf
= {{ .val
= READ_ONCE(skcd
->val
) }};
746 if (sock_cgroup_classid(&skcd_buf
) == classid
)
749 if (!(skcd_buf
.is_data
& 1)) {
751 skcd_buf
.is_data
= 1;
754 skcd_buf
.classid
= classid
;
755 WRITE_ONCE(skcd
->val
, skcd_buf
.val
); /* see sock_cgroup_ptr() */
758 #else /* CONFIG_SOCK_CGROUP_DATA */
760 struct sock_cgroup_data
{
763 #endif /* CONFIG_SOCK_CGROUP_DATA */
765 #endif /* _LINUX_CGROUP_DEFS_H */