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/refcount.h>
17 #include <linux/percpu-refcount.h>
18 #include <linux/percpu-rwsem.h>
19 #include <linux/workqueue.h>
20 #include <linux/bpf-cgroup.h>
27 struct cgroup_taskset;
30 struct kernfs_open_file;
33 #define MAX_CGROUP_TYPE_NAMELEN 32
34 #define MAX_CGROUP_ROOT_NAMELEN 64
35 #define MAX_CFTYPE_NAME 64
37 /* define the enumeration of all cgroup subsystems */
38 #define SUBSYS(_x) _x ## _cgrp_id,
39 enum cgroup_subsys_id {
40 #include <linux/cgroup_subsys.h>
45 /* bits in struct cgroup_subsys_state flags field */
47 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
48 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
49 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
50 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
51 CSS_DYING = (1 << 4), /* css is dying */
54 /* bits in struct cgroup flags field */
56 /* Control Group requires release notifications to userspace */
57 CGRP_NOTIFY_ON_RELEASE,
59 * Clone the parent's configuration when creating a new child
60 * cpuset cgroup. For historical reasons, this option can be
61 * specified at mount time and thus is implemented here.
63 CGRP_CPUSET_CLONE_CHILDREN,
66 /* cgroup_root->flags */
68 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
69 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
72 * Consider namespaces as delegation boundaries. If this flag is
73 * set, controller specific interface files in a namespace root
74 * aren't writeable from inside the namespace.
76 CGRP_ROOT_NS_DELEGATE = (1 << 3),
81 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
82 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
83 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
85 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
86 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
88 /* internal flags, do not use outside cgroup core proper */
89 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
90 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
94 * cgroup_file is the handle for a file instance created in a cgroup which
95 * is used, for example, to generate file changed notifications. This can
96 * be obtained by setting cftype->file_offset.
99 /* do not access any fields from outside cgroup core */
100 struct kernfs_node *kn;
104 * Per-subsystem/per-cgroup state maintained by the system. This is the
105 * fundamental structural building block that controllers deal with.
107 * Fields marked with "PI:" are public and immutable and may be accessed
108 * directly without synchronization.
110 struct cgroup_subsys_state {
111 /* PI: the cgroup that this css is attached to */
112 struct cgroup *cgroup;
114 /* PI: the cgroup subsystem that this css is attached to */
115 struct cgroup_subsys *ss;
117 /* reference count - access via css_[try]get() and css_put() */
118 struct percpu_ref refcnt;
120 /* siblings list anchored at the parent's ->children */
121 struct list_head sibling;
122 struct list_head children;
125 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
126 * matching css can be looked up using css_from_id().
133 * Monotonically increasing unique serial number which defines a
134 * uniform order among all csses. It's guaranteed that all
135 * ->children lists are in the ascending order of ->serial_nr and
136 * used to allow interrupting and resuming iterations.
141 * Incremented by online self and children. Used to guarantee that
142 * parents are not offlined before their children.
146 /* percpu_ref killing and RCU release */
147 struct rcu_head rcu_head;
148 struct work_struct destroy_work;
151 * PI: the parent css. Placed here for cache proximity to following
152 * fields of the containing structure.
154 struct cgroup_subsys_state *parent;
158 * A css_set is a structure holding pointers to a set of
159 * cgroup_subsys_state objects. This saves space in the task struct
160 * object and speeds up fork()/exit(), since a single inc/dec and a
161 * list_add()/del() can bump the reference count on the entire cgroup
166 * Set of subsystem states, one for each subsystem. This array is
167 * immutable after creation apart from the init_css_set during
168 * subsystem registration (at boot time).
170 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
172 /* reference count */
176 * For a domain cgroup, the following points to self. If threaded,
177 * to the matching cset of the nearest domain ancestor. The
178 * dom_cset provides access to the domain cgroup and its csses to
179 * which domain level resource consumptions should be charged.
181 struct css_set *dom_cset;
183 /* the default cgroup associated with this css_set */
184 struct cgroup *dfl_cgrp;
186 /* internal task count, protected by css_set_lock */
190 * Lists running through all tasks using this cgroup group.
191 * mg_tasks lists tasks which belong to this cset but are in the
192 * process of being migrated out or in. Protected by
193 * css_set_rwsem, but, during migration, once tasks are moved to
194 * mg_tasks, it can be read safely while holding cgroup_mutex.
196 struct list_head tasks;
197 struct list_head mg_tasks;
199 /* all css_task_iters currently walking this cset */
200 struct list_head task_iters;
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 threaded csets whose ->dom_cset points to this cset */
212 struct list_head threaded_csets;
213 struct list_head threaded_csets_node;
216 * List running through all cgroup groups in the same hash
217 * slot. Protected by css_set_lock
219 struct hlist_node hlist;
222 * List of cgrp_cset_links pointing at cgroups referenced from this
223 * css_set. Protected by css_set_lock.
225 struct list_head cgrp_links;
228 * List of csets participating in the on-going migration either as
229 * source or destination. Protected by cgroup_mutex.
231 struct list_head mg_preload_node;
232 struct list_head mg_node;
235 * If this cset is acting as the source of migration the following
236 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
237 * respectively the source and destination cgroups of the on-going
238 * migration. mg_dst_cset is the destination cset the target tasks
239 * on this cset should be migrated to. Protected by cgroup_mutex.
241 struct cgroup *mg_src_cgrp;
242 struct cgroup *mg_dst_cgrp;
243 struct css_set *mg_dst_cset;
245 /* dead and being drained, ignore for migration */
248 /* For RCU-protected deletion */
249 struct rcu_head rcu_head;
253 /* self css with NULL ->ss, points back to this cgroup */
254 struct cgroup_subsys_state self;
256 unsigned long flags; /* "unsigned long" so bitops work */
259 * idr allocated in-hierarchy ID.
261 * ID 0 is not used, the ID of the root cgroup is always 1, and a
262 * new cgroup will be assigned with a smallest available ID.
264 * Allocating/Removing ID must be protected by cgroup_mutex.
269 * The depth this cgroup is at. The root is at depth zero and each
270 * step down the hierarchy increments the level. This along with
271 * ancestor_ids[] can determine whether a given cgroup is a
272 * descendant of another without traversing the hierarchy.
276 /* Maximum allowed descent tree depth */
280 * Keep track of total numbers of visible and dying descent cgroups.
281 * Dying cgroups are cgroups which were deleted by a user,
282 * but are still existing because someone else is holding a reference.
283 * max_descendants is a maximum allowed number of descent cgroups.
286 int nr_dying_descendants;
290 * Each non-empty css_set associated with this cgroup contributes
291 * one to nr_populated_csets. The counter is zero iff this cgroup
292 * doesn't have any tasks.
294 * All children which have non-zero nr_populated_csets and/or
295 * nr_populated_children of their own contribute one to either
296 * nr_populated_domain_children or nr_populated_threaded_children
297 * depending on their type. Each counter is zero iff all cgroups
298 * of the type in the subtree proper don't have any tasks.
300 int nr_populated_csets;
301 int nr_populated_domain_children;
302 int nr_populated_threaded_children;
304 int nr_threaded_children; /* # of live threaded child cgroups */
306 struct kernfs_node *kn; /* cgroup kernfs entry */
307 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
308 struct cgroup_file events_file; /* handle for "cgroup.events" */
311 * The bitmask of subsystems enabled on the child cgroups.
312 * ->subtree_control is the one configured through
313 * "cgroup.subtree_control" while ->child_ss_mask is the effective
314 * one which may have more subsystems enabled. Controller knobs
315 * are made available iff it's enabled in ->subtree_control.
319 u16 old_subtree_control;
320 u16 old_subtree_ss_mask;
322 /* Private pointers for each registered subsystem */
323 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
325 struct cgroup_root *root;
328 * List of cgrp_cset_links pointing at css_sets with tasks in this
329 * cgroup. Protected by css_set_lock.
331 struct list_head cset_links;
334 * On the default hierarchy, a css_set for a cgroup with some
335 * susbsys disabled will point to css's which are associated with
336 * the closest ancestor which has the subsys enabled. The
337 * following lists all css_sets which point to this cgroup's css
338 * for the given subsystem.
340 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
343 * If !threaded, self. If threaded, it points to the nearest
344 * domain ancestor. Inside a threaded subtree, cgroups are exempt
345 * from process granularity and no-internal-task constraint.
346 * Domain level resource consumptions which aren't tied to a
347 * specific task are charged to the dom_cgrp.
349 struct cgroup *dom_cgrp;
352 * list of pidlists, up to two for each namespace (one for procs, one
353 * for tasks); created on demand.
355 struct list_head pidlists;
356 struct mutex pidlist_mutex;
358 /* used to wait for offlining of csses */
359 wait_queue_head_t offline_waitq;
361 /* used to schedule release agent */
362 struct work_struct release_agent_work;
364 /* used to store eBPF programs */
365 struct cgroup_bpf bpf;
367 /* ids of the ancestors at each level including self */
372 * A cgroup_root represents the root of a cgroup hierarchy, and may be
373 * associated with a kernfs_root to form an active hierarchy. This is
374 * internal to cgroup core. Don't access directly from controllers.
377 struct kernfs_root *kf_root;
379 /* The bitmask of subsystems attached to this hierarchy */
380 unsigned int subsys_mask;
382 /* Unique id for this hierarchy. */
385 /* The root cgroup. Root is destroyed on its release. */
388 /* for cgrp->ancestor_ids[0] */
389 int cgrp_ancestor_id_storage;
391 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
394 /* A list running through the active hierarchies */
395 struct list_head root_list;
397 /* Hierarchy-specific flags */
400 /* IDs for cgroups in this hierarchy */
401 struct idr cgroup_idr;
403 /* The path to use for release notifications. */
404 char release_agent_path[PATH_MAX];
406 /* The name for this hierarchy - may be empty */
407 char name[MAX_CGROUP_ROOT_NAMELEN];
411 * struct cftype: handler definitions for cgroup control files
413 * When reading/writing to a file:
414 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
415 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
419 * By convention, the name should begin with the name of the
420 * subsystem, followed by a period. Zero length string indicates
421 * end of cftype array.
423 char name[MAX_CFTYPE_NAME];
424 unsigned long private;
427 * The maximum length of string, excluding trailing nul, that can
428 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
430 size_t max_write_len;
436 * If non-zero, should contain the offset from the start of css to
437 * a struct cgroup_file field. cgroup will record the handle of
438 * the created file into it. The recorded handle can be used as
439 * long as the containing css remains accessible.
441 unsigned int file_offset;
444 * Fields used for internal bookkeeping. Initialized automatically
445 * during registration.
447 struct cgroup_subsys *ss; /* NULL for cgroup core files */
448 struct list_head node; /* anchored at ss->cfts */
449 struct kernfs_ops *kf_ops;
451 int (*open)(struct kernfs_open_file *of);
452 void (*release)(struct kernfs_open_file *of);
455 * read_u64() is a shortcut for the common case of returning a
456 * single integer. Use it in place of read()
458 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
460 * read_s64() is a signed version of read_u64()
462 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
464 /* generic seq_file read interface */
465 int (*seq_show)(struct seq_file *sf, void *v);
467 /* optional ops, implement all or none */
468 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
469 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
470 void (*seq_stop)(struct seq_file *sf, void *v);
473 * write_u64() is a shortcut for the common case of accepting
474 * a single integer (as parsed by simple_strtoull) from
475 * userspace. Use in place of write(); return 0 or error.
477 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
480 * write_s64() is a signed version of write_u64()
482 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
486 * write() is the generic write callback which maps directly to
487 * kernfs write operation and overrides all other operations.
488 * Maximum write size is determined by ->max_write_len. Use
489 * of_css/cft() to access the associated css and cft.
491 ssize_t (*write)(struct kernfs_open_file *of,
492 char *buf, size_t nbytes, loff_t off);
494 #ifdef CONFIG_DEBUG_LOCK_ALLOC
495 struct lock_class_key lockdep_key;
500 * Control Group subsystem type.
501 * See Documentation/cgroups/cgroups.txt for details
503 struct cgroup_subsys {
504 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
505 int (*css_online)(struct cgroup_subsys_state *css);
506 void (*css_offline)(struct cgroup_subsys_state *css);
507 void (*css_released)(struct cgroup_subsys_state *css);
508 void (*css_free)(struct cgroup_subsys_state *css);
509 void (*css_reset)(struct cgroup_subsys_state *css);
511 int (*can_attach)(struct cgroup_taskset *tset);
512 void (*cancel_attach)(struct cgroup_taskset *tset);
513 void (*attach)(struct cgroup_taskset *tset);
514 void (*post_attach)(void);
515 int (*can_fork)(struct task_struct *task);
516 void (*cancel_fork)(struct task_struct *task);
517 void (*fork)(struct task_struct *task);
518 void (*exit)(struct task_struct *task);
519 void (*free)(struct task_struct *task);
520 void (*bind)(struct cgroup_subsys_state *root_css);
525 * If %true, the controller, on the default hierarchy, doesn't show
526 * up in "cgroup.controllers" or "cgroup.subtree_control", is
527 * implicitly enabled on all cgroups on the default hierarchy, and
528 * bypasses the "no internal process" constraint. This is for
529 * utility type controllers which is transparent to userland.
531 * An implicit controller can be stolen from the default hierarchy
532 * anytime and thus must be okay with offline csses from previous
533 * hierarchies coexisting with csses for the current one.
535 bool implicit_on_dfl:1;
538 * If %true, the controller, supports threaded mode on the default
539 * hierarchy. In a threaded subtree, both process granularity and
540 * no-internal-process constraint are ignored and a threaded
541 * controllers should be able to handle that.
543 * Note that as an implicit controller is automatically enabled on
544 * all cgroups on the default hierarchy, it should also be
545 * threaded. implicit && !threaded is not supported.
550 * If %false, this subsystem is properly hierarchical -
551 * configuration, resource accounting and restriction on a parent
552 * cgroup cover those of its children. If %true, hierarchy support
553 * is broken in some ways - some subsystems ignore hierarchy
554 * completely while others are only implemented half-way.
556 * It's now disallowed to create nested cgroups if the subsystem is
557 * broken and cgroup core will emit a warning message on such
558 * cases. Eventually, all subsystems will be made properly
559 * hierarchical and this will go away.
561 bool broken_hierarchy:1;
562 bool warned_broken_hierarchy:1;
564 /* the following two fields are initialized automtically during boot */
568 /* optional, initialized automatically during boot if not set */
569 const char *legacy_name;
571 /* link to parent, protected by cgroup_lock() */
572 struct cgroup_root *root;
574 /* idr for css->id */
578 * List of cftypes. Each entry is the first entry of an array
579 * terminated by zero length name.
581 struct list_head cfts;
584 * Base cftypes which are automatically registered. The two can
585 * point to the same array.
587 struct cftype *dfl_cftypes; /* for the default hierarchy */
588 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
591 * A subsystem may depend on other subsystems. When such subsystem
592 * is enabled on a cgroup, the depended-upon subsystems are enabled
593 * together if available. Subsystems enabled due to dependency are
594 * not visible to userland until explicitly enabled. The following
595 * specifies the mask of subsystems that this one depends on.
597 unsigned int depends_on;
600 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
603 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
606 * Allows cgroup operations to synchronize against threadgroup changes
607 * using a percpu_rw_semaphore.
609 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
611 percpu_down_read(&cgroup_threadgroup_rwsem);
615 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
618 * Counterpart of cgroup_threadcgroup_change_begin().
620 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
622 percpu_up_read(&cgroup_threadgroup_rwsem);
625 #else /* CONFIG_CGROUPS */
627 #define CGROUP_SUBSYS_COUNT 0
629 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
634 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
636 #endif /* CONFIG_CGROUPS */
638 #ifdef CONFIG_SOCK_CGROUP_DATA
641 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
642 * per-socket cgroup information except for memcg association.
644 * On legacy hierarchies, net_prio and net_cls controllers directly set
645 * attributes on each sock which can then be tested by the network layer.
646 * On the default hierarchy, each sock is associated with the cgroup it was
647 * created in and the networking layer can match the cgroup directly.
649 * To avoid carrying all three cgroup related fields separately in sock,
650 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
651 * On boot, sock_cgroup_data records the cgroup that the sock was created
652 * in so that cgroup2 matches can be made; however, once either net_prio or
653 * net_cls starts being used, the area is overriden to carry prioidx and/or
654 * classid. The two modes are distinguished by whether the lowest bit is
655 * set. Clear bit indicates cgroup pointer while set bit prioidx and
658 * While userland may start using net_prio or net_cls at any time, once
659 * either is used, cgroup2 matching no longer works. There is no reason to
660 * mix the two and this is in line with how legacy and v2 compatibility is
661 * handled. On mode switch, cgroup references which are already being
662 * pointed to by socks may be leaked. While this can be remedied by adding
663 * synchronization around sock_cgroup_data, given that the number of leaked
664 * cgroups is bound and highly unlikely to be high, this seems to be the
667 struct sock_cgroup_data {
669 #ifdef __LITTLE_ENDIAN
689 * There's a theoretical window where the following accessors race with
690 * updaters and return part of the previous pointer as the prioidx or
691 * classid. Such races are short-lived and the result isn't critical.
693 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
695 /* fallback to 1 which is always the ID of the root cgroup */
696 return (skcd->is_data & 1) ? skcd->prioidx : 1;
699 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
701 /* fallback to 0 which is the unconfigured default classid */
702 return (skcd->is_data & 1) ? skcd->classid : 0;
706 * If invoked concurrently, the updaters may clobber each other. The
707 * caller is responsible for synchronization.
709 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
712 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
714 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
717 if (!(skcd_buf.is_data & 1)) {
719 skcd_buf.is_data = 1;
722 skcd_buf.prioidx = prioidx;
723 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
726 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
729 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
731 if (sock_cgroup_classid(&skcd_buf) == classid)
734 if (!(skcd_buf.is_data & 1)) {
736 skcd_buf.is_data = 1;
739 skcd_buf.classid = classid;
740 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
743 #else /* CONFIG_SOCK_CGROUP_DATA */
745 struct sock_cgroup_data {
748 #endif /* CONFIG_SOCK_CGROUP_DATA */
750 #endif /* _LINUX_CGROUP_DEFS_H */