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>
28 struct cgroup_taskset;
31 struct kernfs_open_file;
34 #define MAX_CGROUP_TYPE_NAMELEN 32
35 #define MAX_CGROUP_ROOT_NAMELEN 64
36 #define MAX_CFTYPE_NAME 64
38 /* define the enumeration of all cgroup subsystems */
39 #define SUBSYS(_x) _x ## _cgrp_id,
40 enum cgroup_subsys_id {
41 #include <linux/cgroup_subsys.h>
46 /* bits in struct cgroup_subsys_state flags field */
48 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
49 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
50 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
51 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
52 CSS_DYING = (1 << 4), /* css is dying */
55 /* bits in struct cgroup flags field */
57 /* Control Group requires release notifications to userspace */
58 CGRP_NOTIFY_ON_RELEASE,
60 * Clone the parent's configuration when creating a new child
61 * cpuset cgroup. For historical reasons, this option can be
62 * specified at mount time and thus is implemented here.
64 CGRP_CPUSET_CLONE_CHILDREN,
67 /* cgroup_root->flags */
69 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
70 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
73 * Consider namespaces as delegation boundaries. If this flag is
74 * set, controller specific interface files in a namespace root
75 * aren't writeable from inside the namespace.
77 CGRP_ROOT_NS_DELEGATE = (1 << 3),
80 * Enable cpuset controller in v1 cgroup to use v2 behavior.
82 CGRP_ROOT_CPUSET_V2_MODE = (1 << 4),
87 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
88 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
89 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
91 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
92 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
94 /* internal flags, do not use outside cgroup core proper */
95 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
96 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
100 * cgroup_file is the handle for a file instance created in a cgroup which
101 * is used, for example, to generate file changed notifications. This can
102 * be obtained by setting cftype->file_offset.
105 /* do not access any fields from outside cgroup core */
106 struct kernfs_node *kn;
110 * Per-subsystem/per-cgroup state maintained by the system. This is the
111 * fundamental structural building block that controllers deal with.
113 * Fields marked with "PI:" are public and immutable and may be accessed
114 * directly without synchronization.
116 struct cgroup_subsys_state {
117 /* PI: the cgroup that this css is attached to */
118 struct cgroup *cgroup;
120 /* PI: the cgroup subsystem that this css is attached to */
121 struct cgroup_subsys *ss;
123 /* reference count - access via css_[try]get() and css_put() */
124 struct percpu_ref refcnt;
126 /* siblings list anchored at the parent's ->children */
127 struct list_head sibling;
128 struct list_head children;
131 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
132 * matching css can be looked up using css_from_id().
139 * Monotonically increasing unique serial number which defines a
140 * uniform order among all csses. It's guaranteed that all
141 * ->children lists are in the ascending order of ->serial_nr and
142 * used to allow interrupting and resuming iterations.
147 * Incremented by online self and children. Used to guarantee that
148 * parents are not offlined before their children.
152 /* percpu_ref killing and RCU release */
153 struct rcu_head rcu_head;
154 struct work_struct destroy_work;
157 * PI: the parent css. Placed here for cache proximity to following
158 * fields of the containing structure.
160 struct cgroup_subsys_state *parent;
164 * A css_set is a structure holding pointers to a set of
165 * cgroup_subsys_state objects. This saves space in the task struct
166 * object and speeds up fork()/exit(), since a single inc/dec and a
167 * list_add()/del() can bump the reference count on the entire cgroup
172 * Set of subsystem states, one for each subsystem. This array is
173 * immutable after creation apart from the init_css_set during
174 * subsystem registration (at boot time).
176 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
178 /* reference count */
182 * For a domain cgroup, the following points to self. If threaded,
183 * to the matching cset of the nearest domain ancestor. The
184 * dom_cset provides access to the domain cgroup and its csses to
185 * which domain level resource consumptions should be charged.
187 struct css_set *dom_cset;
189 /* the default cgroup associated with this css_set */
190 struct cgroup *dfl_cgrp;
192 /* internal task count, protected by css_set_lock */
196 * Lists running through all tasks using this cgroup group.
197 * mg_tasks lists tasks which belong to this cset but are in the
198 * process of being migrated out or in. Protected by
199 * css_set_rwsem, but, during migration, once tasks are moved to
200 * mg_tasks, it can be read safely while holding cgroup_mutex.
202 struct list_head tasks;
203 struct list_head mg_tasks;
204 struct list_head dying_tasks;
206 /* all css_task_iters currently walking this cset */
207 struct list_head task_iters;
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 threaded csets whose ->dom_cset points to this cset */
219 struct list_head threaded_csets;
220 struct list_head threaded_csets_node;
223 * List running through all cgroup groups in the same hash
224 * slot. Protected by css_set_lock
226 struct hlist_node hlist;
229 * List of cgrp_cset_links pointing at cgroups referenced from this
230 * css_set. Protected by css_set_lock.
232 struct list_head cgrp_links;
235 * List of csets participating in the on-going migration either as
236 * source or destination. Protected by cgroup_mutex.
238 struct list_head mg_src_preload_node;
239 struct list_head mg_dst_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.
293 * nr_descendants and nr_dying_descendants are protected
294 * by cgroup_mutex and css_set_lock. It's fine to read them holding
295 * any of cgroup_mutex and css_set_lock; for writing both locks
299 int nr_dying_descendants;
303 * Each non-empty css_set associated with this cgroup contributes
304 * one to nr_populated_csets. The counter is zero iff this cgroup
305 * doesn't have any tasks.
307 * All children which have non-zero nr_populated_csets and/or
308 * nr_populated_children of their own contribute one to either
309 * nr_populated_domain_children or nr_populated_threaded_children
310 * depending on their type. Each counter is zero iff all cgroups
311 * of the type in the subtree proper don't have any tasks.
313 int nr_populated_csets;
314 int nr_populated_domain_children;
315 int nr_populated_threaded_children;
317 int nr_threaded_children; /* # of live threaded child cgroups */
319 struct kernfs_node *kn; /* cgroup kernfs entry */
320 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
321 struct cgroup_file events_file; /* handle for "cgroup.events" */
324 * The bitmask of subsystems enabled on the child cgroups.
325 * ->subtree_control is the one configured through
326 * "cgroup.subtree_control" while ->child_ss_mask is the effective
327 * one which may have more subsystems enabled. Controller knobs
328 * are made available iff it's enabled in ->subtree_control.
332 u16 old_subtree_control;
333 u16 old_subtree_ss_mask;
335 /* Private pointers for each registered subsystem */
336 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
338 struct cgroup_root *root;
341 * List of cgrp_cset_links pointing at css_sets with tasks in this
342 * cgroup. Protected by css_set_lock.
344 struct list_head cset_links;
347 * On the default hierarchy, a css_set for a cgroup with some
348 * susbsys disabled will point to css's which are associated with
349 * the closest ancestor which has the subsys enabled. The
350 * following lists all css_sets which point to this cgroup's css
351 * for the given subsystem.
353 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
356 * If !threaded, self. If threaded, it points to the nearest
357 * domain ancestor. Inside a threaded subtree, cgroups are exempt
358 * from process granularity and no-internal-task constraint.
359 * Domain level resource consumptions which aren't tied to a
360 * specific task are charged to the dom_cgrp.
362 struct cgroup *dom_cgrp;
363 struct cgroup *old_dom_cgrp; /* used while enabling threaded */
366 * list of pidlists, up to two for each namespace (one for procs, one
367 * for tasks); created on demand.
369 struct list_head pidlists;
370 struct mutex pidlist_mutex;
372 /* used to wait for offlining of csses */
373 wait_queue_head_t offline_waitq;
375 /* used to schedule release agent */
376 struct work_struct release_agent_work;
378 /* used to store eBPF programs */
379 struct cgroup_bpf bpf;
381 /* ids of the ancestors at each level including self */
386 * A cgroup_root represents the root of a cgroup hierarchy, and may be
387 * associated with a kernfs_root to form an active hierarchy. This is
388 * internal to cgroup core. Don't access directly from controllers.
391 struct kernfs_root *kf_root;
393 /* The bitmask of subsystems attached to this hierarchy */
394 unsigned int subsys_mask;
396 /* Unique id for this hierarchy. */
399 /* The root cgroup. Root is destroyed on its release. */
402 /* for cgrp->ancestor_ids[0] */
403 int cgrp_ancestor_id_storage;
405 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
408 /* A list running through the active hierarchies */
409 struct list_head root_list;
411 /* Hierarchy-specific flags */
414 /* IDs for cgroups in this hierarchy */
415 struct idr cgroup_idr;
417 /* The path to use for release notifications. */
418 char release_agent_path[PATH_MAX];
420 /* The name for this hierarchy - may be empty */
421 char name[MAX_CGROUP_ROOT_NAMELEN];
425 * struct cftype: handler definitions for cgroup control files
427 * When reading/writing to a file:
428 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
429 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
433 * By convention, the name should begin with the name of the
434 * subsystem, followed by a period. Zero length string indicates
435 * end of cftype array.
437 char name[MAX_CFTYPE_NAME];
438 unsigned long private;
441 * The maximum length of string, excluding trailing nul, that can
442 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
444 size_t max_write_len;
450 * If non-zero, should contain the offset from the start of css to
451 * a struct cgroup_file field. cgroup will record the handle of
452 * the created file into it. The recorded handle can be used as
453 * long as the containing css remains accessible.
455 unsigned int file_offset;
458 * Fields used for internal bookkeeping. Initialized automatically
459 * during registration.
461 struct cgroup_subsys *ss; /* NULL for cgroup core files */
462 struct list_head node; /* anchored at ss->cfts */
463 struct kernfs_ops *kf_ops;
465 int (*open)(struct kernfs_open_file *of);
466 void (*release)(struct kernfs_open_file *of);
469 * read_u64() is a shortcut for the common case of returning a
470 * single integer. Use it in place of read()
472 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
474 * read_s64() is a signed version of read_u64()
476 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
478 /* generic seq_file read interface */
479 int (*seq_show)(struct seq_file *sf, void *v);
481 /* optional ops, implement all or none */
482 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
483 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
484 void (*seq_stop)(struct seq_file *sf, void *v);
487 * write_u64() is a shortcut for the common case of accepting
488 * a single integer (as parsed by simple_strtoull) from
489 * userspace. Use in place of write(); return 0 or error.
491 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
494 * write_s64() is a signed version of write_u64()
496 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
500 * write() is the generic write callback which maps directly to
501 * kernfs write operation and overrides all other operations.
502 * Maximum write size is determined by ->max_write_len. Use
503 * of_css/cft() to access the associated css and cft.
505 ssize_t (*write)(struct kernfs_open_file *of,
506 char *buf, size_t nbytes, loff_t off);
508 #ifdef CONFIG_DEBUG_LOCK_ALLOC
509 struct lock_class_key lockdep_key;
514 * Control Group subsystem type.
515 * See Documentation/cgroups/cgroups.txt for details
517 struct cgroup_subsys {
518 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
519 int (*css_online)(struct cgroup_subsys_state *css);
520 void (*css_offline)(struct cgroup_subsys_state *css);
521 void (*css_released)(struct cgroup_subsys_state *css);
522 void (*css_free)(struct cgroup_subsys_state *css);
523 void (*css_reset)(struct cgroup_subsys_state *css);
525 int (*can_attach)(struct cgroup_taskset *tset);
526 void (*cancel_attach)(struct cgroup_taskset *tset);
527 void (*attach)(struct cgroup_taskset *tset);
528 void (*post_attach)(void);
529 int (*can_fork)(struct task_struct *task);
530 void (*cancel_fork)(struct task_struct *task);
531 void (*fork)(struct task_struct *task);
532 void (*exit)(struct task_struct *task);
533 void (*release)(struct task_struct *task);
534 void (*bind)(struct cgroup_subsys_state *root_css);
539 * If %true, the controller, on the default hierarchy, doesn't show
540 * up in "cgroup.controllers" or "cgroup.subtree_control", is
541 * implicitly enabled on all cgroups on the default hierarchy, and
542 * bypasses the "no internal process" constraint. This is for
543 * utility type controllers which is transparent to userland.
545 * An implicit controller can be stolen from the default hierarchy
546 * anytime and thus must be okay with offline csses from previous
547 * hierarchies coexisting with csses for the current one.
549 bool implicit_on_dfl:1;
552 * If %true, the controller, supports threaded mode on the default
553 * hierarchy. In a threaded subtree, both process granularity and
554 * no-internal-process constraint are ignored and a threaded
555 * controllers should be able to handle that.
557 * Note that as an implicit controller is automatically enabled on
558 * all cgroups on the default hierarchy, it should also be
559 * threaded. implicit && !threaded is not supported.
564 * If %false, this subsystem is properly hierarchical -
565 * configuration, resource accounting and restriction on a parent
566 * cgroup cover those of its children. If %true, hierarchy support
567 * is broken in some ways - some subsystems ignore hierarchy
568 * completely while others are only implemented half-way.
570 * It's now disallowed to create nested cgroups if the subsystem is
571 * broken and cgroup core will emit a warning message on such
572 * cases. Eventually, all subsystems will be made properly
573 * hierarchical and this will go away.
575 bool broken_hierarchy:1;
576 bool warned_broken_hierarchy:1;
578 /* the following two fields are initialized automtically during boot */
582 /* optional, initialized automatically during boot if not set */
583 const char *legacy_name;
585 /* link to parent, protected by cgroup_lock() */
586 struct cgroup_root *root;
588 /* idr for css->id */
592 * List of cftypes. Each entry is the first entry of an array
593 * terminated by zero length name.
595 struct list_head cfts;
598 * Base cftypes which are automatically registered. The two can
599 * point to the same array.
601 struct cftype *dfl_cftypes; /* for the default hierarchy */
602 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
605 * A subsystem may depend on other subsystems. When such subsystem
606 * is enabled on a cgroup, the depended-upon subsystems are enabled
607 * together if available. Subsystems enabled due to dependency are
608 * not visible to userland until explicitly enabled. The following
609 * specifies the mask of subsystems that this one depends on.
611 unsigned int depends_on;
614 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
617 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
620 * Allows cgroup operations to synchronize against threadgroup changes
621 * using a percpu_rw_semaphore.
623 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
625 percpu_down_read(&cgroup_threadgroup_rwsem);
629 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
632 * Counterpart of cgroup_threadcgroup_change_begin().
634 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
636 percpu_up_read(&cgroup_threadgroup_rwsem);
639 #else /* CONFIG_CGROUPS */
641 #define CGROUP_SUBSYS_COUNT 0
643 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
648 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
650 #endif /* CONFIG_CGROUPS */
652 #ifdef CONFIG_SOCK_CGROUP_DATA
655 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
656 * per-socket cgroup information except for memcg association.
658 * On legacy hierarchies, net_prio and net_cls controllers directly set
659 * attributes on each sock which can then be tested by the network layer.
660 * On the default hierarchy, each sock is associated with the cgroup it was
661 * created in and the networking layer can match the cgroup directly.
663 * To avoid carrying all three cgroup related fields separately in sock,
664 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
665 * On boot, sock_cgroup_data records the cgroup that the sock was created
666 * in so that cgroup2 matches can be made; however, once either net_prio or
667 * net_cls starts being used, the area is overriden to carry prioidx and/or
668 * classid. The two modes are distinguished by whether the lowest bit is
669 * set. Clear bit indicates cgroup pointer while set bit prioidx and
672 * While userland may start using net_prio or net_cls at any time, once
673 * either is used, cgroup2 matching no longer works. There is no reason to
674 * mix the two and this is in line with how legacy and v2 compatibility is
675 * handled. On mode switch, cgroup references which are already being
676 * pointed to by socks may be leaked. While this can be remedied by adding
677 * synchronization around sock_cgroup_data, given that the number of leaked
678 * cgroups is bound and highly unlikely to be high, this seems to be the
681 struct sock_cgroup_data {
683 #ifdef __LITTLE_ENDIAN
707 * There's a theoretical window where the following accessors race with
708 * updaters and return part of the previous pointer as the prioidx or
709 * classid. Such races are short-lived and the result isn't critical.
711 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
713 /* fallback to 1 which is always the ID of the root cgroup */
714 return (skcd->is_data & 1) ? skcd->prioidx : 1;
717 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
719 /* fallback to 0 which is the unconfigured default classid */
720 return (skcd->is_data & 1) ? skcd->classid : 0;
724 * If invoked concurrently, the updaters may clobber each other. The
725 * caller is responsible for synchronization.
727 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
730 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
732 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
735 if (!(skcd_buf.is_data & 1)) {
737 skcd_buf.is_data = 1;
740 skcd_buf.prioidx = prioidx;
741 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
744 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
747 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
749 if (sock_cgroup_classid(&skcd_buf) == classid)
752 if (!(skcd_buf.is_data & 1)) {
754 skcd_buf.is_data = 1;
757 skcd_buf.classid = classid;
758 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
761 #else /* CONFIG_SOCK_CGROUP_DATA */
763 struct sock_cgroup_data {
766 #endif /* CONFIG_SOCK_CGROUP_DATA */
768 #endif /* _LINUX_CGROUP_DEFS_H */