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>
19 #include <linux/bpf-cgroup.h>
26 struct cgroup_taskset;
29 struct kernfs_open_file;
32 #define MAX_CGROUP_TYPE_NAMELEN 32
33 #define MAX_CGROUP_ROOT_NAMELEN 64
34 #define MAX_CFTYPE_NAME 64
36 /* define the enumeration of all cgroup subsystems */
37 #define SUBSYS(_x) _x ## _cgrp_id,
38 enum cgroup_subsys_id {
39 #include <linux/cgroup_subsys.h>
44 /* bits in struct cgroup_subsys_state flags field */
46 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
47 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
48 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
49 CSS_VISIBLE = (1 << 3), /* css is visible to userland */
52 /* bits in struct cgroup flags field */
54 /* Control Group requires release notifications to userspace */
55 CGRP_NOTIFY_ON_RELEASE,
57 * Clone the parent's configuration when creating a new child
58 * cpuset cgroup. For historical reasons, this option can be
59 * specified at mount time and thus is implemented here.
61 CGRP_CPUSET_CLONE_CHILDREN,
64 /* cgroup_root->flags */
66 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
67 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
72 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
73 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
74 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
75 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
77 /* internal flags, do not use outside cgroup core proper */
78 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
79 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
83 * cgroup_file is the handle for a file instance created in a cgroup which
84 * is used, for example, to generate file changed notifications. This can
85 * be obtained by setting cftype->file_offset.
88 /* do not access any fields from outside cgroup core */
89 struct kernfs_node *kn;
93 * Per-subsystem/per-cgroup state maintained by the system. This is the
94 * fundamental structural building block that controllers deal with.
96 * Fields marked with "PI:" are public and immutable and may be accessed
97 * directly without synchronization.
99 struct cgroup_subsys_state {
100 /* PI: the cgroup that this css is attached to */
101 struct cgroup *cgroup;
103 /* PI: the cgroup subsystem that this css is attached to */
104 struct cgroup_subsys *ss;
106 /* reference count - access via css_[try]get() and css_put() */
107 struct percpu_ref refcnt;
109 /* PI: the parent css */
110 struct cgroup_subsys_state *parent;
112 /* siblings list anchored at the parent's ->children */
113 struct list_head sibling;
114 struct list_head children;
117 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
118 * matching css can be looked up using css_from_id().
125 * Monotonically increasing unique serial number which defines a
126 * uniform order among all csses. It's guaranteed that all
127 * ->children lists are in the ascending order of ->serial_nr and
128 * used to allow interrupting and resuming iterations.
133 * Incremented by online self and children. Used to guarantee that
134 * parents are not offlined before their children.
138 /* percpu_ref killing and RCU release */
139 struct rcu_head rcu_head;
140 struct work_struct destroy_work;
144 * A css_set is a structure holding pointers to a set of
145 * cgroup_subsys_state objects. This saves space in the task struct
146 * object and speeds up fork()/exit(), since a single inc/dec and a
147 * list_add()/del() can bump the reference count on the entire cgroup
152 * Set of subsystem states, one for each subsystem. This array is
153 * immutable after creation apart from the init_css_set during
154 * subsystem registration (at boot time).
156 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
158 /* reference count */
161 /* the default cgroup associated with this css_set */
162 struct cgroup *dfl_cgrp;
165 * Lists running through all tasks using this cgroup group.
166 * mg_tasks lists tasks which belong to this cset but are in the
167 * process of being migrated out or in. Protected by
168 * css_set_rwsem, but, during migration, once tasks are moved to
169 * mg_tasks, it can be read safely while holding cgroup_mutex.
171 struct list_head tasks;
172 struct list_head mg_tasks;
174 /* all css_task_iters currently walking this cset */
175 struct list_head task_iters;
178 * On the default hierarhcy, ->subsys[ssid] may point to a css
179 * attached to an ancestor instead of the cgroup this css_set is
180 * associated with. The following node is anchored at
181 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
182 * iterate through all css's attached to a given cgroup.
184 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
187 * List running through all cgroup groups in the same hash
188 * slot. Protected by css_set_lock
190 struct hlist_node hlist;
193 * List of cgrp_cset_links pointing at cgroups referenced from this
194 * css_set. Protected by css_set_lock.
196 struct list_head cgrp_links;
199 * List of csets participating in the on-going migration either as
200 * source or destination. Protected by cgroup_mutex.
202 struct list_head mg_preload_node;
203 struct list_head mg_node;
206 * If this cset is acting as the source of migration the following
207 * two fields are set. mg_src_cgrp and mg_dst_cgrp are
208 * respectively the source and destination cgroups of the on-going
209 * migration. mg_dst_cset is the destination cset the target tasks
210 * on this cset should be migrated to. Protected by cgroup_mutex.
212 struct cgroup *mg_src_cgrp;
213 struct cgroup *mg_dst_cgrp;
214 struct css_set *mg_dst_cset;
216 /* dead and being drained, ignore for migration */
219 /* For RCU-protected deletion */
220 struct rcu_head rcu_head;
224 /* self css with NULL ->ss, points back to this cgroup */
225 struct cgroup_subsys_state self;
227 unsigned long flags; /* "unsigned long" so bitops work */
230 * idr allocated in-hierarchy ID.
232 * ID 0 is not used, the ID of the root cgroup is always 1, and a
233 * new cgroup will be assigned with a smallest available ID.
235 * Allocating/Removing ID must be protected by cgroup_mutex.
240 * The depth this cgroup is at. The root is at depth zero and each
241 * step down the hierarchy increments the level. This along with
242 * ancestor_ids[] can determine whether a given cgroup is a
243 * descendant of another without traversing the hierarchy.
248 * Each non-empty css_set associated with this cgroup contributes
249 * one to populated_cnt. All children with non-zero popuplated_cnt
250 * of their own contribute one. The count is zero iff there's no
251 * task in this cgroup or its subtree.
255 struct kernfs_node *kn; /* cgroup kernfs entry */
256 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
257 struct cgroup_file events_file; /* handle for "cgroup.events" */
260 * The bitmask of subsystems enabled on the child cgroups.
261 * ->subtree_control is the one configured through
262 * "cgroup.subtree_control" while ->child_ss_mask is the effective
263 * one which may have more subsystems enabled. Controller knobs
264 * are made available iff it's enabled in ->subtree_control.
268 u16 old_subtree_control;
269 u16 old_subtree_ss_mask;
271 /* Private pointers for each registered subsystem */
272 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
274 struct cgroup_root *root;
277 * List of cgrp_cset_links pointing at css_sets with tasks in this
278 * cgroup. Protected by css_set_lock.
280 struct list_head cset_links;
283 * On the default hierarchy, a css_set for a cgroup with some
284 * susbsys disabled will point to css's which are associated with
285 * the closest ancestor which has the subsys enabled. The
286 * following lists all css_sets which point to this cgroup's css
287 * for the given subsystem.
289 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
292 * list of pidlists, up to two for each namespace (one for procs, one
293 * for tasks); created on demand.
295 struct list_head pidlists;
296 struct mutex pidlist_mutex;
298 /* used to wait for offlining of csses */
299 wait_queue_head_t offline_waitq;
301 /* used to schedule release agent */
302 struct work_struct release_agent_work;
304 /* used to store eBPF programs */
305 struct cgroup_bpf bpf;
307 /* ids of the ancestors at each level including self */
312 * A cgroup_root represents the root of a cgroup hierarchy, and may be
313 * associated with a kernfs_root to form an active hierarchy. This is
314 * internal to cgroup core. Don't access directly from controllers.
317 struct kernfs_root *kf_root;
319 /* The bitmask of subsystems attached to this hierarchy */
320 unsigned int subsys_mask;
322 /* Unique id for this hierarchy. */
325 /* The root cgroup. Root is destroyed on its release. */
328 /* for cgrp->ancestor_ids[0] */
329 int cgrp_ancestor_id_storage;
331 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
334 /* A list running through the active hierarchies */
335 struct list_head root_list;
337 /* Hierarchy-specific flags */
340 /* IDs for cgroups in this hierarchy */
341 struct idr cgroup_idr;
343 /* The path to use for release notifications. */
344 char release_agent_path[PATH_MAX];
346 /* The name for this hierarchy - may be empty */
347 char name[MAX_CGROUP_ROOT_NAMELEN];
351 * struct cftype: handler definitions for cgroup control files
353 * When reading/writing to a file:
354 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
355 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
359 * By convention, the name should begin with the name of the
360 * subsystem, followed by a period. Zero length string indicates
361 * end of cftype array.
363 char name[MAX_CFTYPE_NAME];
364 unsigned long private;
367 * The maximum length of string, excluding trailing nul, that can
368 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
370 size_t max_write_len;
376 * If non-zero, should contain the offset from the start of css to
377 * a struct cgroup_file field. cgroup will record the handle of
378 * the created file into it. The recorded handle can be used as
379 * long as the containing css remains accessible.
381 unsigned int file_offset;
384 * Fields used for internal bookkeeping. Initialized automatically
385 * during registration.
387 struct cgroup_subsys *ss; /* NULL for cgroup core files */
388 struct list_head node; /* anchored at ss->cfts */
389 struct kernfs_ops *kf_ops;
391 int (*open)(struct kernfs_open_file *of);
392 void (*release)(struct kernfs_open_file *of);
395 * read_u64() is a shortcut for the common case of returning a
396 * single integer. Use it in place of read()
398 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
400 * read_s64() is a signed version of read_u64()
402 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
404 /* generic seq_file read interface */
405 int (*seq_show)(struct seq_file *sf, void *v);
407 /* optional ops, implement all or none */
408 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
409 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
410 void (*seq_stop)(struct seq_file *sf, void *v);
413 * write_u64() is a shortcut for the common case of accepting
414 * a single integer (as parsed by simple_strtoull) from
415 * userspace. Use in place of write(); return 0 or error.
417 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
420 * write_s64() is a signed version of write_u64()
422 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
426 * write() is the generic write callback which maps directly to
427 * kernfs write operation and overrides all other operations.
428 * Maximum write size is determined by ->max_write_len. Use
429 * of_css/cft() to access the associated css and cft.
431 ssize_t (*write)(struct kernfs_open_file *of,
432 char *buf, size_t nbytes, loff_t off);
434 #ifdef CONFIG_DEBUG_LOCK_ALLOC
435 struct lock_class_key lockdep_key;
440 * Control Group subsystem type.
441 * See Documentation/cgroups/cgroups.txt for details
443 struct cgroup_subsys {
444 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
445 int (*css_online)(struct cgroup_subsys_state *css);
446 void (*css_offline)(struct cgroup_subsys_state *css);
447 void (*css_released)(struct cgroup_subsys_state *css);
448 void (*css_free)(struct cgroup_subsys_state *css);
449 void (*css_reset)(struct cgroup_subsys_state *css);
451 int (*can_attach)(struct cgroup_taskset *tset);
452 void (*cancel_attach)(struct cgroup_taskset *tset);
453 void (*attach)(struct cgroup_taskset *tset);
454 void (*post_attach)(void);
455 int (*can_fork)(struct task_struct *task);
456 void (*cancel_fork)(struct task_struct *task);
457 void (*fork)(struct task_struct *task);
458 void (*exit)(struct task_struct *task);
459 void (*free)(struct task_struct *task);
460 void (*bind)(struct cgroup_subsys_state *root_css);
465 * If %true, the controller, on the default hierarchy, doesn't show
466 * up in "cgroup.controllers" or "cgroup.subtree_control", is
467 * implicitly enabled on all cgroups on the default hierarchy, and
468 * bypasses the "no internal process" constraint. This is for
469 * utility type controllers which is transparent to userland.
471 * An implicit controller can be stolen from the default hierarchy
472 * anytime and thus must be okay with offline csses from previous
473 * hierarchies coexisting with csses for the current one.
475 bool implicit_on_dfl:1;
478 * If %false, this subsystem is properly hierarchical -
479 * configuration, resource accounting and restriction on a parent
480 * cgroup cover those of its children. If %true, hierarchy support
481 * is broken in some ways - some subsystems ignore hierarchy
482 * completely while others are only implemented half-way.
484 * It's now disallowed to create nested cgroups if the subsystem is
485 * broken and cgroup core will emit a warning message on such
486 * cases. Eventually, all subsystems will be made properly
487 * hierarchical and this will go away.
489 bool broken_hierarchy:1;
490 bool warned_broken_hierarchy:1;
492 /* the following two fields are initialized automtically during boot */
496 /* optional, initialized automatically during boot if not set */
497 const char *legacy_name;
499 /* link to parent, protected by cgroup_lock() */
500 struct cgroup_root *root;
502 /* idr for css->id */
506 * List of cftypes. Each entry is the first entry of an array
507 * terminated by zero length name.
509 struct list_head cfts;
512 * Base cftypes which are automatically registered. The two can
513 * point to the same array.
515 struct cftype *dfl_cftypes; /* for the default hierarchy */
516 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
519 * A subsystem may depend on other subsystems. When such subsystem
520 * is enabled on a cgroup, the depended-upon subsystems are enabled
521 * together if available. Subsystems enabled due to dependency are
522 * not visible to userland until explicitly enabled. The following
523 * specifies the mask of subsystems that this one depends on.
525 unsigned int depends_on;
528 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
531 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
534 * Allows cgroup operations to synchronize against threadgroup changes
535 * using a percpu_rw_semaphore.
537 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
539 percpu_down_read(&cgroup_threadgroup_rwsem);
543 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
546 * Counterpart of cgroup_threadcgroup_change_begin().
548 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
550 percpu_up_read(&cgroup_threadgroup_rwsem);
553 #else /* CONFIG_CGROUPS */
555 #define CGROUP_SUBSYS_COUNT 0
557 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
562 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
564 #endif /* CONFIG_CGROUPS */
566 #ifdef CONFIG_SOCK_CGROUP_DATA
569 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
570 * per-socket cgroup information except for memcg association.
572 * On legacy hierarchies, net_prio and net_cls controllers directly set
573 * attributes on each sock which can then be tested by the network layer.
574 * On the default hierarchy, each sock is associated with the cgroup it was
575 * created in and the networking layer can match the cgroup directly.
577 * To avoid carrying all three cgroup related fields separately in sock,
578 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
579 * On boot, sock_cgroup_data records the cgroup that the sock was created
580 * in so that cgroup2 matches can be made; however, once either net_prio or
581 * net_cls starts being used, the area is overriden to carry prioidx and/or
582 * classid. The two modes are distinguished by whether the lowest bit is
583 * set. Clear bit indicates cgroup pointer while set bit prioidx and
586 * While userland may start using net_prio or net_cls at any time, once
587 * either is used, cgroup2 matching no longer works. There is no reason to
588 * mix the two and this is in line with how legacy and v2 compatibility is
589 * handled. On mode switch, cgroup references which are already being
590 * pointed to by socks may be leaked. While this can be remedied by adding
591 * synchronization around sock_cgroup_data, given that the number of leaked
592 * cgroups is bound and highly unlikely to be high, this seems to be the
595 struct sock_cgroup_data {
597 #ifdef __LITTLE_ENDIAN
617 * There's a theoretical window where the following accessors race with
618 * updaters and return part of the previous pointer as the prioidx or
619 * classid. Such races are short-lived and the result isn't critical.
621 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
623 /* fallback to 1 which is always the ID of the root cgroup */
624 return (skcd->is_data & 1) ? skcd->prioidx : 1;
627 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
629 /* fallback to 0 which is the unconfigured default classid */
630 return (skcd->is_data & 1) ? skcd->classid : 0;
634 * If invoked concurrently, the updaters may clobber each other. The
635 * caller is responsible for synchronization.
637 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
640 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
642 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
645 if (!(skcd_buf.is_data & 1)) {
647 skcd_buf.is_data = 1;
650 skcd_buf.prioidx = prioidx;
651 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
654 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
657 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
659 if (sock_cgroup_classid(&skcd_buf) == classid)
662 if (!(skcd_buf.is_data & 1)) {
664 skcd_buf.is_data = 1;
667 skcd_buf.classid = classid;
668 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
671 #else /* CONFIG_SOCK_CGROUP_DATA */
673 struct sock_cgroup_data {
676 #endif /* CONFIG_SOCK_CGROUP_DATA */
678 #endif /* _LINUX_CGROUP_DEFS_H */