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 */
74 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
75 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
76 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
77 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
79 /* internal flags, do not use outside cgroup core proper */
80 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
81 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
85 * cgroup_file is the handle for a file instance created in a cgroup which
86 * is used, for example, to generate file changed notifications. This can
87 * be obtained by setting cftype->file_offset.
90 /* do not access any fields from outside cgroup core */
91 struct kernfs_node *kn;
95 * Per-subsystem/per-cgroup state maintained by the system. This is the
96 * fundamental structural building block that controllers deal with.
98 * Fields marked with "PI:" are public and immutable and may be accessed
99 * directly without synchronization.
101 struct cgroup_subsys_state {
102 /* PI: the cgroup that this css is attached to */
103 struct cgroup *cgroup;
105 /* PI: the cgroup subsystem that this css is attached to */
106 struct cgroup_subsys *ss;
108 /* reference count - access via css_[try]get() and css_put() */
109 struct percpu_ref refcnt;
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;
142 * PI: the parent css. Placed here for cache proximity to following
143 * fields of the containing structure.
145 struct cgroup_subsys_state *parent;
149 * A css_set is a structure holding pointers to a set of
150 * cgroup_subsys_state objects. This saves space in the task struct
151 * object and speeds up fork()/exit(), since a single inc/dec and a
152 * list_add()/del() can bump the reference count on the entire cgroup
157 * Set of subsystem states, one for each subsystem. This array is
158 * immutable after creation apart from the init_css_set during
159 * subsystem registration (at boot time).
161 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
163 /* reference count */
166 /* the default cgroup associated with this css_set */
167 struct cgroup *dfl_cgrp;
170 * Lists running through all tasks using this cgroup group.
171 * mg_tasks lists tasks which belong to this cset but are in the
172 * process of being migrated out or in. Protected by
173 * css_set_rwsem, but, during migration, once tasks are moved to
174 * mg_tasks, it can be read safely while holding cgroup_mutex.
176 struct list_head tasks;
177 struct list_head mg_tasks;
179 /* all css_task_iters currently walking this cset */
180 struct list_head task_iters;
183 * On the default hierarhcy, ->subsys[ssid] may point to a css
184 * attached to an ancestor instead of the cgroup this css_set is
185 * associated with. The following node is anchored at
186 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
187 * iterate through all css's attached to a given cgroup.
189 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
192 * List running through all cgroup groups in the same hash
193 * slot. Protected by css_set_lock
195 struct hlist_node hlist;
198 * List of cgrp_cset_links pointing at cgroups referenced from this
199 * css_set. Protected by css_set_lock.
201 struct list_head cgrp_links;
204 * List of csets participating in the on-going migration either as
205 * source or destination. Protected by cgroup_mutex.
207 struct list_head mg_preload_node;
208 struct list_head mg_node;
211 * If this cset is acting as the source of migration the following
212 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
213 * respectively the source and destination cgroups of the on-going
214 * migration. mg_dst_cset is the destination cset the target tasks
215 * on this cset should be migrated to. Protected by cgroup_mutex.
217 struct cgroup *mg_src_cgrp;
218 struct cgroup *mg_dst_cgrp;
219 struct css_set *mg_dst_cset;
221 /* dead and being drained, ignore for migration */
224 /* For RCU-protected deletion */
225 struct rcu_head rcu_head;
229 /* self css with NULL ->ss, points back to this cgroup */
230 struct cgroup_subsys_state self;
232 unsigned long flags; /* "unsigned long" so bitops work */
235 * idr allocated in-hierarchy ID.
237 * ID 0 is not used, the ID of the root cgroup is always 1, and a
238 * new cgroup will be assigned with a smallest available ID.
240 * Allocating/Removing ID must be protected by cgroup_mutex.
245 * The depth this cgroup is at. The root is at depth zero and each
246 * step down the hierarchy increments the level. This along with
247 * ancestor_ids[] can determine whether a given cgroup is a
248 * descendant of another without traversing the hierarchy.
253 * Each non-empty css_set associated with this cgroup contributes
254 * one to populated_cnt. All children with non-zero popuplated_cnt
255 * of their own contribute one. The count is zero iff there's no
256 * task in this cgroup or its subtree.
260 struct kernfs_node *kn; /* cgroup kernfs entry */
261 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
262 struct cgroup_file events_file; /* handle for "cgroup.events" */
265 * The bitmask of subsystems enabled on the child cgroups.
266 * ->subtree_control is the one configured through
267 * "cgroup.subtree_control" while ->child_ss_mask is the effective
268 * one which may have more subsystems enabled. Controller knobs
269 * are made available iff it's enabled in ->subtree_control.
273 u16 old_subtree_control;
274 u16 old_subtree_ss_mask;
276 /* Private pointers for each registered subsystem */
277 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
279 struct cgroup_root *root;
282 * List of cgrp_cset_links pointing at css_sets with tasks in this
283 * cgroup. Protected by css_set_lock.
285 struct list_head cset_links;
288 * On the default hierarchy, a css_set for a cgroup with some
289 * susbsys disabled will point to css's which are associated with
290 * the closest ancestor which has the subsys enabled. The
291 * following lists all css_sets which point to this cgroup's css
292 * for the given subsystem.
294 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
297 * list of pidlists, up to two for each namespace (one for procs, one
298 * for tasks); created on demand.
300 struct list_head pidlists;
301 struct mutex pidlist_mutex;
303 /* used to wait for offlining of csses */
304 wait_queue_head_t offline_waitq;
306 /* used to schedule release agent */
307 struct work_struct release_agent_work;
309 /* used to store eBPF programs */
310 struct cgroup_bpf bpf;
312 /* ids of the ancestors at each level including self */
317 * A cgroup_root represents the root of a cgroup hierarchy, and may be
318 * associated with a kernfs_root to form an active hierarchy. This is
319 * internal to cgroup core. Don't access directly from controllers.
322 struct kernfs_root *kf_root;
324 /* The bitmask of subsystems attached to this hierarchy */
325 unsigned int subsys_mask;
327 /* Unique id for this hierarchy. */
330 /* The root cgroup. Root is destroyed on its release. */
333 /* for cgrp->ancestor_ids[0] */
334 int cgrp_ancestor_id_storage;
336 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
339 /* A list running through the active hierarchies */
340 struct list_head root_list;
342 /* Hierarchy-specific flags */
345 /* IDs for cgroups in this hierarchy */
346 struct idr cgroup_idr;
348 /* The path to use for release notifications. */
349 char release_agent_path[PATH_MAX];
351 /* The name for this hierarchy - may be empty */
352 char name[MAX_CGROUP_ROOT_NAMELEN];
356 * struct cftype: handler definitions for cgroup control files
358 * When reading/writing to a file:
359 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
360 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
364 * By convention, the name should begin with the name of the
365 * subsystem, followed by a period. Zero length string indicates
366 * end of cftype array.
368 char name[MAX_CFTYPE_NAME];
369 unsigned long private;
372 * The maximum length of string, excluding trailing nul, that can
373 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
375 size_t max_write_len;
381 * If non-zero, should contain the offset from the start of css to
382 * a struct cgroup_file field. cgroup will record the handle of
383 * the created file into it. The recorded handle can be used as
384 * long as the containing css remains accessible.
386 unsigned int file_offset;
389 * Fields used for internal bookkeeping. Initialized automatically
390 * during registration.
392 struct cgroup_subsys *ss; /* NULL for cgroup core files */
393 struct list_head node; /* anchored at ss->cfts */
394 struct kernfs_ops *kf_ops;
396 int (*open)(struct kernfs_open_file *of);
397 void (*release)(struct kernfs_open_file *of);
400 * read_u64() is a shortcut for the common case of returning a
401 * single integer. Use it in place of read()
403 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
405 * read_s64() is a signed version of read_u64()
407 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
409 /* generic seq_file read interface */
410 int (*seq_show)(struct seq_file *sf, void *v);
412 /* optional ops, implement all or none */
413 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
414 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
415 void (*seq_stop)(struct seq_file *sf, void *v);
418 * write_u64() is a shortcut for the common case of accepting
419 * a single integer (as parsed by simple_strtoull) from
420 * userspace. Use in place of write(); return 0 or error.
422 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
425 * write_s64() is a signed version of write_u64()
427 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
431 * write() is the generic write callback which maps directly to
432 * kernfs write operation and overrides all other operations.
433 * Maximum write size is determined by ->max_write_len. Use
434 * of_css/cft() to access the associated css and cft.
436 ssize_t (*write)(struct kernfs_open_file *of,
437 char *buf, size_t nbytes, loff_t off);
439 #ifdef CONFIG_DEBUG_LOCK_ALLOC
440 struct lock_class_key lockdep_key;
445 * Control Group subsystem type.
446 * See Documentation/cgroups/cgroups.txt for details
448 struct cgroup_subsys {
449 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
450 int (*css_online)(struct cgroup_subsys_state *css);
451 void (*css_offline)(struct cgroup_subsys_state *css);
452 void (*css_released)(struct cgroup_subsys_state *css);
453 void (*css_free)(struct cgroup_subsys_state *css);
454 void (*css_reset)(struct cgroup_subsys_state *css);
456 int (*can_attach)(struct cgroup_taskset *tset);
457 void (*cancel_attach)(struct cgroup_taskset *tset);
458 void (*attach)(struct cgroup_taskset *tset);
459 void (*post_attach)(void);
460 int (*can_fork)(struct task_struct *task);
461 void (*cancel_fork)(struct task_struct *task);
462 void (*fork)(struct task_struct *task);
463 void (*exit)(struct task_struct *task);
464 void (*free)(struct task_struct *task);
465 void (*bind)(struct cgroup_subsys_state *root_css);
470 * If %true, the controller, on the default hierarchy, doesn't show
471 * up in "cgroup.controllers" or "cgroup.subtree_control", is
472 * implicitly enabled on all cgroups on the default hierarchy, and
473 * bypasses the "no internal process" constraint. This is for
474 * utility type controllers which is transparent to userland.
476 * An implicit controller can be stolen from the default hierarchy
477 * anytime and thus must be okay with offline csses from previous
478 * hierarchies coexisting with csses for the current one.
480 bool implicit_on_dfl:1;
483 * If %false, this subsystem is properly hierarchical -
484 * configuration, resource accounting and restriction on a parent
485 * cgroup cover those of its children. If %true, hierarchy support
486 * is broken in some ways - some subsystems ignore hierarchy
487 * completely while others are only implemented half-way.
489 * It's now disallowed to create nested cgroups if the subsystem is
490 * broken and cgroup core will emit a warning message on such
491 * cases. Eventually, all subsystems will be made properly
492 * hierarchical and this will go away.
494 bool broken_hierarchy:1;
495 bool warned_broken_hierarchy:1;
497 /* the following two fields are initialized automtically during boot */
501 /* optional, initialized automatically during boot if not set */
502 const char *legacy_name;
504 /* link to parent, protected by cgroup_lock() */
505 struct cgroup_root *root;
507 /* idr for css->id */
511 * List of cftypes. Each entry is the first entry of an array
512 * terminated by zero length name.
514 struct list_head cfts;
517 * Base cftypes which are automatically registered. The two can
518 * point to the same array.
520 struct cftype *dfl_cftypes; /* for the default hierarchy */
521 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
524 * A subsystem may depend on other subsystems. When such subsystem
525 * is enabled on a cgroup, the depended-upon subsystems are enabled
526 * together if available. Subsystems enabled due to dependency are
527 * not visible to userland until explicitly enabled. The following
528 * specifies the mask of subsystems that this one depends on.
530 unsigned int depends_on;
533 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
536 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
539 * Allows cgroup operations to synchronize against threadgroup changes
540 * using a percpu_rw_semaphore.
542 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
544 percpu_down_read(&cgroup_threadgroup_rwsem);
548 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
551 * Counterpart of cgroup_threadcgroup_change_begin().
553 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
555 percpu_up_read(&cgroup_threadgroup_rwsem);
558 #else /* CONFIG_CGROUPS */
560 #define CGROUP_SUBSYS_COUNT 0
562 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
567 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
569 #endif /* CONFIG_CGROUPS */
571 #ifdef CONFIG_SOCK_CGROUP_DATA
574 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
575 * per-socket cgroup information except for memcg association.
577 * On legacy hierarchies, net_prio and net_cls controllers directly set
578 * attributes on each sock which can then be tested by the network layer.
579 * On the default hierarchy, each sock is associated with the cgroup it was
580 * created in and the networking layer can match the cgroup directly.
582 * To avoid carrying all three cgroup related fields separately in sock,
583 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
584 * On boot, sock_cgroup_data records the cgroup that the sock was created
585 * in so that cgroup2 matches can be made; however, once either net_prio or
586 * net_cls starts being used, the area is overriden to carry prioidx and/or
587 * classid. The two modes are distinguished by whether the lowest bit is
588 * set. Clear bit indicates cgroup pointer while set bit prioidx and
591 * While userland may start using net_prio or net_cls at any time, once
592 * either is used, cgroup2 matching no longer works. There is no reason to
593 * mix the two and this is in line with how legacy and v2 compatibility is
594 * handled. On mode switch, cgroup references which are already being
595 * pointed to by socks may be leaked. While this can be remedied by adding
596 * synchronization around sock_cgroup_data, given that the number of leaked
597 * cgroups is bound and highly unlikely to be high, this seems to be the
600 struct sock_cgroup_data {
602 #ifdef __LITTLE_ENDIAN
622 * There's a theoretical window where the following accessors race with
623 * updaters and return part of the previous pointer as the prioidx or
624 * classid. Such races are short-lived and the result isn't critical.
626 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
628 /* fallback to 1 which is always the ID of the root cgroup */
629 return (skcd->is_data & 1) ? skcd->prioidx : 1;
632 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
634 /* fallback to 0 which is the unconfigured default classid */
635 return (skcd->is_data & 1) ? skcd->classid : 0;
639 * If invoked concurrently, the updaters may clobber each other. The
640 * caller is responsible for synchronization.
642 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
645 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
647 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
650 if (!(skcd_buf.is_data & 1)) {
652 skcd_buf.is_data = 1;
655 skcd_buf.prioidx = prioidx;
656 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
659 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
662 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
664 if (sock_cgroup_classid(&skcd_buf) == classid)
667 if (!(skcd_buf.is_data & 1)) {
669 skcd_buf.is_data = 1;
672 skcd_buf.classid = classid;
673 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
676 #else /* CONFIG_SOCK_CGROUP_DATA */
678 struct sock_cgroup_data {
681 #endif /* CONFIG_SOCK_CGROUP_DATA */
683 #endif /* _LINUX_CGROUP_DEFS_H */